Skip to main content
Download PDF

Artificial intelligence: illuminating the depths of the tumor microenvironment

Journal of Translational Medicine volume 22, Article number: 799 (2024) Cite this article

Abstract

Artificial intelligence (AI) can acquire characteristics that are not yet known to humans through extensive learning, enabling to handle large amounts of pathology image data. Divided into machine learning and deep learning, AI has the advantage of handling large amounts of data and processing image analysis, consequently it also has a great potential in accurately assessing tumour microenvironment (TME) models. With the complex composition of the TME, in-depth study of TME contributes to new ideas for treatment, assessment of patient response to postoperative therapy and prognostic prediction. This leads to a review of the development of AI's application in TME assessment in this study, provides an overview of AI techniques applied to medicine, delves into the application of AI in analysing the quantitative and spatial location characteristics of various cells (tumour cells, immune and non-immune cells) in the TME, reveals the predictive prognostic value of TME and provides new ideas for tumour therapy, highlights the great potential for clinical applications. In addition, a discussion of its limitations and encouraging future directions for its practical clinical application is presented.

Introduction

The extracellular matrix, a plethora of reactive chemicals, immune cells including lymphocytes, macrophages, and neutrophils, together with non-immune cells like fibroblasts and vascular endothelial cells, make up the highly complex components of the tumor microenvironment (TME). The TME exerts a crucial part in tumour occurrence, growth, prognosis and metastasis [1]. Characterisation of the tumour microenvironment as a predictor of a patient's prognosis, and also affects the sensitivity of the tumour to treatment, amongst other things, enabling doctors to assess a patient's disease progression and survival, and to develop a personalised treatment plan for the patient.

Current pathologists use histopathology-based microscopy to evaluate TME along with the quantification and localization of cells therein, which are prone to the risk of sample bias and subjectivity. Additionally, standard methodologies are unable to swiftly and intuitively extract the multi-dimensional features of the tumor microenvironment due to its multi-dimensional features, which encompass both quantitative and spatial characteristics of various parameters. Single-cell genomics, spatial transcriptomics, multiplex immunofluorescence and other analytical methods are applied to the study of TME [2]. Nevertheless, these approaches are costly, labor-intensive, time-consuming when dealing with large amounts of data that can only be obtained from a single data source (e.g., gene expression, cellular images), as well as insufficiently tapping into the large number of cellular interactions to completely capture the diversity and dynamics in the TME, hampering a thorough study of spatial aspects specific to the TME [3,4,5]. Therefore, it is necessary to apply more precise, convenient, and objective analytical methods to assess TME. Artificial intelligence (AI) has been used to digital pathology images for use in activities linked to cancer diagnosis, prognosis, and prediction. AI has the ability to integrate data from different sources, such as genomics, transcriptomics, imaging, etc., to provide more comprehensive information about the TME. When it comes to handling large data sets, AI also has a major advantage. By analyzing TME-related data quickly and efficiently, AI is able to discover hidden biological features, disease mechanisms or potential therapeutic targets in TME. Deep learning (DL) and machine learning (ML) are two aspects of AI. DL and ML techniques have become powerful tools to evaluate TME, for instance, it is used to study the interaction and number of immune cells and tumour-associated cells in TME for observing the impact on patient prognosis. AI learns the spatial location of each cell in the TME through supervised learning methods, so as to further analyses whether cells in various locations have varied relevance in the TME [6].

Drawing briefly on the application of deep learning and machine learning to pathological images based on hematoxylin and eosin (H&E) staining, this paper focuses on the study of the application of AI in analysing the quantity and spatial location of TME and its cellular components, mainly in terms of tumour cells, immune cells (TILs, TAMs, TANs), and non-immune cells (CAFs). It also highlights the considerable advantages that AI has in analysing TME, and the integration of AI with technologies such as spatial transcriptomics in the future will enable more precise access to cellular interactions and positional relationships, as well as reveal differences in the expression and spatial distribution of genes at the level of different regions within the tissues, different cell types, and even individual cells, which will help to explore the mechanisms of diseases, discover potential in TME therapeutic targets, etc.

Overview of artificial intelligence

AI is the computer system's simulation of human intelligence processes. By leveraging large-scale datasets, AI models learn intricate patterns and features, surpassing traditional methods in detecting subtle morphological changes indicative of various cancers. ML, a branch of AI that uses statistical techniques to optimize task-specific models [7]. Predictive models can be constructed by extracting information related to patient prognosis from tumour pathology images. DL, while one of the most advanced ML methods, applies neural networks to learn deep patterns in image data, which can enhance the analysis of images [8]. AI improves the digitisation of pathology with the capability of effectively identifying tissue biological features on pathology slides. Numerous pathological image processing and classification activities, such as tumor classification, grading, prognosis prediction, and treatment, can also be accomplished with it. In addition to minimizing diagnostic errors brought on by pathologists' technical differences and conserving diagnostic time, AI makes it possible for pathology analysis to go from qualitative to quantitative analysis.

Machine learning

ML is the process of building predictive models by using labelled training set data, identifying and extracting features, applying the learned rules to new data and making predictions or decisions without the need for explicit programming. There are routine steps of data preparation, model selection, model training, model evaluation, parameter tuning and prediction in machine learning [9]. According to the training method, ML can be divided into three main categories, supervised learning, unsupervised learning [10], and reinforcement learning. Currently ML applied to pathology images is usually supervised learning, which requires professional pathologists to annotate the images before allowing the ML model to train the data for the further development of the prediction model. By integrating genomic, transcriptomic, proteomic, and metabolomic data, ML is able to reveal the complex interactions in the TME, which is now the latest application of ML in TME [11].

In machine learning methods such as Support Vector Machines (SVM) and Random Forests (RF), normally the features of most importance for tumor development and treatment response are extracted manually for modeling and classification, which reduces the data dimensions and improves the efficiency of the analysis [12]. However, pathological images often exhibit significant variations, requiring strong expertise for feature extraction, which can be incomplete and thus lead to lower classification accuracy.

Deep learning

Compared with machine learning, deep learning makes it possible to overcome the limitations of manual feature extraction and automatically extract complex nonlinear features from data, which has been gradually and widely used in the classification of pathology images [13]. The application of DL algorithms to pathology images is expected to change the way malignant tumour pathology is diagnosed and stratified for treatment, and is another milestone event in the application of AI in medicine. DL frameworks build on the proposal that neural networks acquire representations and computations [14] similar to those of the biological brain by learning sample data to automatically determine the intrinsic regularities and levels of representation existed in the features from the input data.

Types of deep learning models

Convolutional neural network

Predominantly composed of convolutional, pooling, and fully connected layers, Convolutional Neural Networks (CNNs) are currently the most widely used deep learning algorithm for digital pathology image analysis [15]. CNNs are commonly used for pathology image analysis and visual feature extraction of tumor tissues to identify tumor regions and cell types. As can be seen in the Fig. 1, a CNN-based deep learning model extracts feature and performs learning and classification by performing convolutional operations and pooling operations on input data. Firstly, the WSIs are disassembled into small patches. Secondly, preprocess the data such as staining normalization and data enhancement. Lastly, construct an AI-based model. The processed data can be divided into training set and validation set, which are used for training optimization of CNN models. The trained models can be deployed to new data for testing to evaluate model performance. CNN has already been employed for detection and segmentation tasks of pathological images with the ability to be used to identify and quantify cells on the one hand and classify them on the other hand. For instance, it is possible to sort out various cells in the TME such as neutrophils and lymphocytes at the cellular level [16], and also separate tumour from non-tumour regions, grade the malignancy of tumours, and so on.

Fig. 1
figure 1

Typical process of CNN-based approach for pathology image analysis

Full size image

Recurrent neural networks

The application of recurrent neural networks (RNNs) in medical image analysis is not very widespread than several other deep learning algorithms, instead it is often applied to text analysis or natural language processing [17]. RNNs are neural network model types specifically designed to cope with sequential data, which are also capable of capturing temporal information in medical images such as the response to treatment in tumor patients over time, or the temporal dynamics of processing gene expression data. Unlike CNNs, RNNs feature the ability to process image or numerical data and analyse tissue images obtained at distinct stages. As depicted in Fig. 2, inputs are based on HE-stained WSI, with the RNN model outputting classification results by combining channel attention and spatial attention, which further predict 5-year survival [18]. There are numerous RNNs malformed networks, with the Long Short-Term Network (LSTM) being one of the most widely utilized networks. Targeted keywords that help pathologists write pathology reports can be obtained by combining LSTM with pathology picture recognition. A neural network model combining LSTM and CNN was deployed by Bychkov et al. [18] to predict the five-year survival rate of colorectal cancer patients using HE-stained pathology images. Research results indicated that the model's prediction accuracy was significantly better than other classifiers and higher than that of a visual risk score model.

Fig. 2
figure 2

The RNN model outputs classification results and predicts 5-year survival

Full size image

Generative adversarial network

The basic structure of Generative adversarial networks (GAN) includes a generator and a discriminator, which continuously optimize the loss function through their adversarial interplay to generate pseudo data highly similar to real data. GANs enable to be used for generating virtual tumor image data, for training data-poor models, or for simulating tumor behavior under different microenvironmental conditions. Therefore, GANs are often used for color normalization in pathological images to reduce the impact of color on classification, just as illustrated in Fig. 3. Moreover, extensive research applies GANs to virtual staining of pathological images, showing potential clinical applications [19]. In order to perform virtual immunohistochemistry (IHC) stained on the same slide, Xu et al. [20] employed a GAN network to transform HE-stained digital pathology images into IHC-stained images. This approach eliminated the negative consequences of destructive IHC-based tissue testing while simultaneously improving experimental efficiency due to the little amount of manual labelling data required. Although virtual staining through GAN provides cost reduction, safety, etc., the use of it for clinical use is currently immature and requires standardization of staining, improved robustness of staining results, etc. In addition to applying GAN in the field of pathology images, there have been studies using it for magnetic resonance (MR) image processing. For instance, an adversarial learning framework for multimodal MR image fusion was trained and validated on the glioma dataset by Liu and his team [21], and the results revealed that the method outperforms some of the latest techniques in medical image fusion.

Fig. 3
figure 3

GAN model is capable of stain normalization and virtual IHC staining

Full size image

Transformers

Originally proposed by Vaswani et al. [22] in 2017, the Transformer model has become a groundbreaking technique that utilizes a "self-attention" mechanism to capture the intrinsic relationships in the input data without relying on traditional RNN or CNN structures, it contains mainly decoders and encoders. Expected results it delivers are better than other models, and it parallelizes training, which is fast and solves the problem of long-distance dependence well, with the exception that it is based entirely on self-attention, with a certain amount of loss of information about location. As shown in Fig. 4, the transformer model accomplishes the task of tumor segmentation based on HE-stained pathology images by means of a decoder, an encoder, and a multilayer perceptron used for feature transformations, with classification of the benign and malignant tumor. Subsequently, TransUNet (Transformers and U-Net) was presented, which opened up the application of Transformer in the field of medical image segmentation. The fusion of Transformer and U-Net minimizes the amount of computation, and it provides a better advantage in large-scale datasets and captures important information effectively.

Fig. 4
figure 4

Transformer model to segment pathology images and classify the benign and malignant of tumors

Full size image

AI for multidimensional characterization of TME

The phenotype and function of cells in TME may highly depend on the precise spatial location of cells and their interactions with neighboring cells. Therefore, Accurate cell segmentation and classification are necessary to analyze the multidimensional spatial characteristics of TME. Currently, the assessment of TME by histological methods is prone to sample errors due to the difficulties associated with obtaining high-quality tissue sections, as well as the spatial heterogeneity within the tumor and the dynamic evolution of TME. Utilizing machine learning algorithms to process large-scale tumor tissue section images and single-cell data, it is able to construct models of cell types and spatial distribution in the tumor microenvironment. In addition, extracting features in tumor tissue sections, such as morphological and spatial distribution features, AI is able to perform cell type classification and spatial distribution analysis of TME [11].

Application of a single model to characterize TME

As opposed to traditional AI models for segmenting tumour images, Zhu et al. [23] developed a CNN-based brain tumour segmentation model in their recent study, which consisted of three modules combining multimodal, spatial and boundary information to analyse the global spatiality of the image, which facilitated the accurate acquisition of the tumour's location and interrelationships with other tissues on magnetic resonance images. The model was validated by external datasets with superiority in both performance and computational efficiency. Nagy et al. [24] improved the MultiOmyx analysis process based on a DL model developed in the NeoGenomics lab. This DL model produced biomarker intensities, phenotype counts, phenotype densities, and cellular morphological information in addition to segmenting and classifying images. In addition to this, it was able to perform advanced spatial analyses to pinpoint the clustering patterns of different phenotypes, which contributed to the investigation of complex cellular interactions in TME.

Application of fusion models to characterize TME

With the integration of learning, deep learning fusion models can combine the strengths of multiple single models to significantly improve predictive performance, enhance model robustness and stability, and support complex data and tasks at the same time.

In order to comprehensively analyze the spatial features of tumors, Liu et al. [25] combined the stronger local information extraction capability of CNNs and the excellent global representation capability of Transformers to build a hybrid model named TransSea for the task of brain tumor segmentation in medical images. By training and testing the BraTS2020 and BraTS2021 datasets, TransSea obtained Dice scores of 86.32% and 90.84%, respectively, which is a clear advantage over other models.

Although spatial transcriptomics is capable of in-depth analysis of the relationship between tumors and TME, it is costly and has limitations in practical clinical use. Based on this, Gao et al. [26] developed a deep learning model based on CNN and GAN (IGI-DL model), which is capable of predicting the expression of spatial transcriptomics in patients based on H&E-stained histological images by learning pixel intensities and structural features, effectively reducing the technical cost of using spatial transcriptomics. IGI-DL was also able to characterize the spatial features of TME, determine the heterogeneity of TME, and demonstrate that TME plays an important role in cancer prognosis. Comparative analysis with other models (e.g., HisToGene, DeepSpaCE, etc.) showed that IGI-DL exhibits optimal performance in predicting 179 target genes, both in the test and validation sets. Although the model currently performed well in only three cancer types, its ability to characterize TME could provide an effective bridge for probing spatial gene expression.

AI quantification of cells within the TME

Recent advancements in DL and ML techniques, have revolutionized the field of pathology by enabling precise cell identification, detection, quantification, and localization as well as identifying subtle changes in gene expression, metabolite levels or protein structure associated with disease [27].

Deep learning techniques based on segmentation of pathology images at the cellular level

DL techniques applied to segmentation of cellular level pathology images are U-Net [28], DeepCell [29], CellProfiler [30] and so on. With little cellular annotation, U-net based on CNN is able to rely on data augmentation to improve the robustness and invariance of the model [31]. As one of the pioneers in the field of AI-driven cellular analysis, DeepCell began by identifying cell populations based on morphology alone, and later was able to identify intracellular heterogeneity based on subtle morphological differences, its continuous development has provided an excellent platform for biological experiments at cellular level as well as for medical research. Owing to its ability to accurately differentiate between various immune cell subtypes, various cancer cells, and stromal cells, it is capable of being used for cellular profiling, cell and gene therapy development, and stem cell research, among others. On the basis of machine learning, CellProfiler [32] is able to automate the analysis of individual cells, quickly and accurately measuring various characteristics of the cells, such as size, shape, brightness, and so on.

Furthermore, the potential of machine learning to evaluate the TME is highlighted by the application of supervised machine learning to digital images of HE-stained tissue microarrays by Väyrynen et al. [33], which classified and counted lymphocytes, plasma cells, neutrophils and eosinophils in intra-epithelial and mesenchymal zones of colorectal cancer tumours. It's not hard to conclude that artificial intelligence-based analysis of WSIs will accelerate pathologists' assessment of the complex TME and increase the objectivity and reproducibility of predictions.

AI in tumor cells

The technologies currently used to study the complexity and heterogeneity of cells in TME are predominantly single-cell sequencing technologies, flow cytometry and others. In contrast, AI approaches, such as image analysis of pathology slides, can offer insights into spatial relationships between different cell types and their distribution within the TME. Moreover, AI algorithms trained on large datasets can potentially identify and characterize rare cell types or subtle phenotypic changes with greater accuracy and efficiency [34], as illustrated in Table 1.

Table 1 Application of AI for assessment of cells in TME of different tumors
Full size table

For the identification and capture of tumour cells, the application of CNN is preferable. In order to further optimise the cell detection and classification function of the VGGNet model, Li et al. [35] designed a CNN model with flow cytometry-derived datasets, which was able to achieve precise capture of cancer cells in a few milliseconds with more than 95% accuracy. In addition, for tumour cell classification, deep learning is able to extract features and achieve high accuracy in the classification of unlabelled cells. To distinguish cancer cells derived from cholangiocarcinoma within an unlabeled microscopy image, Chawan et al. [36] developed a proof-of-concept deep learning model by morphological differences. However, distinguishing cells by morphology alone can mistakenly miss identifying broken cells, cells with abnormally large morphology, and non-cellular objects that resemble cells, and whose accuracy therefore needs to be reconsidered. In addition, in order to differentiate between benign and malignant urothelial cells, Masatomo Kaneko et al. [34] developed a CNN model including the EfficientNet B6 and Arcface architectures which successfully differentiated between all cellular subtypes of urothelial cells, achieving up to 90% accuracy.

AI in immune cells

Tumor-infiltrating lymphocytes

Tumour-infiltrating lymphocytes (TILs) have been shown to be tumour-killing and exert an essential effect in the identification of tumour antigens [37, 38]. TILs contain both positively regulating immune response immune cells, such as CD4 T cells, CD8 T cells, NK cells, Th1 cells, and Tfh cells, which are capable of suppressing tumors [39, 40]. Conversely, myeloid suppressor (MDSC), Treg cells, Th2 cells, etc. are able to promote tumor growth, as is shown in the Fig. 5A.

Fig. 5
figure 5

The complex tumor microenvironment. A Tumour-infiltrating lymphocytes. B Tumour-associated macrophage. C Tumor-associated neutrophils. D Cancer-associated fibroblasts

Full size image

The degree of TILs’ infiltration within the tumour is usually positively correlated with the efficacy of immune checkpoint inhibitors, with higher levels of infiltration being associated with better efficacy and prognosis [41]. The number, type, and region of TILs within tumour tissues are important in predicting solid tumour clinical prognosis [42]. TILs can be used as a predictor of higher pCR rates with neoadjuvant chemotherapy, As shown in Table 2. One of the studies had analyzed 498 patients with HER2-positive breast cancer treated with neoadjuvant treatment [43]. The results noted that TILs contribute to the prediction and prognosis of these patients.

Table 2 Prognostic value of different cells in the TME
Full size table

For the purpose of providing standardized and effective TIL quantification, automated image analysis approaches particularly AI-based methods are required, which offer standardized criteria for stringent validation by qualified pathologists and quality control by regulatory bodies [44,45,46]. Just as Table 1 illustrates, they also increase quantitative accuracy, reduce time, and make it easier to analyze more complicated spatial patterns. Joel et al. [47] developed a comprehensive approach and an interactive tool that incorporated expert feedback into a deep learning model based on extensive previous research, which could accurately generate TIL Maps from WSIs. This iterative feedback increased the overall accuracy of the TIL Maps. Both the necrosis segmentation CNN and the lymphocyte infiltration categorization CNN were applied. The first one distinguished between the little areas of the input image that had lymphocyte infiltration and those that did not. Initialized with an unsupervised convolutional autoencoder (CAE), it was a semi-supervised CNN. In order to reduce false positives in necrotic zones—where cell nuclei may resemble regions invaded by lymphocytes—the latter segmented the necrotic sections. The study revealed that the degree of TILs penetration may influence overall survival as well as the spatial aspects of the TME. Juha et al. [33] performed image analysis of HE-stained slides of the TME of CRC patients by a ML based approach to identify four types of immune cells in the TME: neutrophils, eosinophils, plasma cells, and other lymphocytes, which were subsequently classified. Results of the training showed that this automated approach to detect and classify immune cells using machine learning was highly consistent with pathologists and independently trained automated classifiers. The findings also revealed that high density of lymphocytes and eosinophils was known to be correlated with better survival.

Tumour-associated macrophages

As the most diverse immune cell in the TME: the tumour-associated macrophage (TAM), commonly linked to poor prognosis and drug resistance, is classified into two distinct subtypes according to morphological, phenotypic, and functional heterogeneity, namely the M1 and the M2 types [48, 49]. The two subtypes play diametrically opposed roles in the TME [50, 51]. As is depicted in the Fig. 5B, M1 TAMs with anticancer effects can release pro-inflammatory mediators such as IL-1, IL-12, IL-18, IL-23, and TNFα. M2 TAMs are triggered by IL4 and M-CSF [52]. M2 TAMs exhibit high levels of expression for several factors involved in cell adhesion and proliferation, including insulin-like growth factor (IGF), platelet-derived growth factor (PDGF), betaig-h3 (BIG-H3), and fibronectin (FN) [53, 54]. TAMs are frequently linked to a poor clinical prognosis in cancer patients [48, 55, 56]. However, more recent research has shown that the prognostic significance of TAMs is debatable and that the positional distribution and function of TAMs affect a tumor's prognosis. As indicated in the Table 2, according to Li et al., a poor prognosis was closely linked to the accumulation of CD163 TAMs in lung cancer [55]. Interestingly, several clinical studies support the value of counting TAMs for prognostic and predictive outcomes. For example, an in-depth study performed by Ruffell et al. inticated that lymph node metastasis and inadequate pathological staging in patients with breast cancer were linked to macrophage infiltration (CD68+) [57]. Macrophages were not found in breast tissue in patients who did not receive chemotherapy; instead, they were more common in normal tissues that were not adjacent. In contrast, the tumor infiltrating macrophage levels were higher in patients undergoing neoadjuvant chemotherapy.

TAMs play a key role in TME and tumour biology, and few studies have applied AI to TAMs. On the one hand, it is because the development of AI and precision medicine needs to be further improved, and on the other hand, it is because TAMs are heterogeneous and their complexity makes the combination of AI and TAMs challenging [58]. Classification and medical picture segmentation are the main areas of application for neural networks and deep learning in TAMs. A recent study developed a deep learning-based computational model, Mask R-CNN, for segmenting cell nuclei from HE-stained pathological images of lung adenocarcinoma, which included segmentation and classification of macrophage nuclei [59]. Additionally, using machine learning techniques in the Orange Data Mining Toolbox, researchers were able to create a quick and easy imaging-based approach that could recognize various macrophage functional phenotypes based on cell size and morphology [60]. This machine learning approach, which solely examined macrophage morphology, demonstrated 90% average accuracy in identifying M1 and M2 phenotypes and differentiating them from naïve macrophages and monocytes. Random forest (RF) is an ML algorithm that ranks each variable's predictive potential and builds predictive models. It is a supervised learning technique based on feature stochastic vectors [61]. In 2022, in an attempt to investigate the prognostic significance of macrophages and their heterogeneous phenotypes in non-small cell lung cancer, Wu et al. [62] screened for prognostic markers using a machine learning algorithm with a RF model and constructed an immune-related risk score based on CD68 to predict disease-free survival.

To elucidate the regulatory role of macrophage infiltration in high-grade plasmacytoid ovarian cancer (HGSOC), Chang et al. developed a macrophage-associated predictive model utilizing the ML LASSO method and validated it in different HGSOC cohorts [63]. The results showed that high levels of M1 TAMs infiltration were related with favorable outcomes, but high levels of M2 TAMs infiltration were related with bad outcomes. Shen et al. [64] created a DL model to define the immune infiltration of the transcriptome with the goal of classifying brain tumours according to their distinct immune infiltration characteristics. To handle gene expression data, the model made use of an eighteen-layer ResNet feature encoder. The feature encoder was trained using close to 100,000 transcriptomes from various cancer types. The model identified two molecular subtypes, C1 and C2, of brain tumors, each with a distinct immunological infiltration profile and prognosis. It was determined that the considerable TAM infiltration of the C2 subtype was a key characteristic.

Tumour-associated neutrophils

Analogously to TAMs, tumor-associated neutrophils (TANs) are fall into two distinct phenotypes: Anti-tumor phenotype N1 TANs and pro-tumor phenotype N2 TANs. As can be seen in the Fig. 5C, N1 TANs directly produce cytotoxic mediators such as reactive oxygen species (ROS) and myeloperoxidase (MPO). In the presence of interferon β (IFN-β), the N1 TANs are suppressed and then converted to the N2 TANs. N2 TANs promote tumor invasion and angiogenesis by producing cytokines such as matrix metalloproteinase 9 (MMP9), vascular endothelial growth factor (VEGF), and hepatocyte growth factor (HGF). In addition, neutrophils produce H2O2, CCL3, CXCL9, CXCL10, ROS, NETs and Arg1 to regress tumour.

As shown in the Table 2, the prognosis resulting from differences in the relative location of TAN versus tumour cells varies, as well as the results of several studies support a strong correlation between intratumoural neutrophils and poor prognosis and a weak correlation between peritumoural and mesenchymal neutrophils and poor prognosis [65]. Increased neutrophils have been linked to a worse prognosis in various malignancies, including glioma, metastatic melanoma, and gastrointestinal mesenchymal tumors, according to a number of studies conducted over the past few decades. Recently, an in-depth study performed by Chen et al. [66] explored the plasticity of N1 TANs and N2 TANs in TME of PDAC and the effect of their immune infiltration on the prognostic value of patients. A total of 119 patients undergoing radical resection were included in this study, and N1 TANs and N2 TANs were identified by immunofluorescence staining, and the plasticity of N1 and N2 was evaluated by the N1/N2 ratio. Multivariate factor analysis showed that a low N1/N2 ratio was associated with poorer tumour differentiation, milder lymph node metastasis and higher TNM stage. While this was not the case for N2 TANs, the group with a large number of N1 TANs had a significantly longer median OS and RFS than the group with a low number of N1 TANs.

It is shown in Table 1, using image processing technology and ML algorithms, the quantity of TANs in cancer tissue can be evaluated rapidly and precisely, providing a significant basis for tumour progression and prognosis assessment [67, 68]. Using computational simulation and machine learning technologies, the interaction process between TANs and tumour cells can be simulated to gain an in-depth understanding of their mechanism of action. Artificial intelligence-based immunosurveillance technology can monitor the dynamic changes of TANs in the TME in real time, providing real-time guidance for clinical treatment. In the area of drug development, combined with AI technology, TANs can be used as targets for drug screening and discovery of new tumour therapeutic drugs [69, 70].

Some researchers proposed a deep learning model for identifying myeloproliferative tumours based on neutrophil morphology construction, and chose the PicoDet deep learning target detection method to determine whether a patient is a myeloproliferative tumour and myeloproliferative tumour subtype [71]. The outcomes demonstrated that PicoDet can identify the cells in the bone marrow smear more accurately and achieve four classifications of the dataset, whose average accuracy rates were all over 70%, achieving a good classification prediction effect. Bi et al. [72] subdivided 6115 neutrophils from the WSI of malignant hematological diseases, trained these neutrophils using a migration learning algorithm, built a convolutional neural network model based on the morphological phenotypes of the neutrophils to determine their disease classification, and evaluated the model using confusion matrices and subject arithmetic characteristic (ROC) curves. The results showed that neutrophils from various diseases could be categorized into distinct groups, and the accuracy of the DL model in judging neutrophils of different diseases reached 0.896. Differences in neutrophil nuclear morphology may underlie the heterogeneity. Therefore, some researchers have conducted a preliminary study of neutrophil phenotypic heterogeneity in different haematological diseases by deep learning [73]. Firstly, neutrophil images were manually segmented and then nuclei were segmented using the interactive semantic segmentation tool ilastik. Validation results showed that the model achieved an accuracy of 0.749. The study also analyzed the nuclear features of neutrophils through migration learning, a machine learning-based pixel classification technique.

Cancer-associated fibroblasts

Among non-immune cells, cancer-associated fibroblasts (CAF) are the most numerous of the TMEs, accounting for about 80%, with the involvement in tumour generation, growth and drug resistance, consequently they are considered to be pro-tumourigenic [74]. Through their interactions with other TME components, CAFs play a crucial role in shaping TMEs, indicating their potential utility as therapeutic targets and prognostic variables. CAFs are heterogeneous, in that CAFs also exert pleiotropic functions in TMEs. MyCAF, apCAF, iCAF, and vCAF are the four primary subtypes of CAF [75, 76]. By producing cytokines such as TGFβ, IL6 and IL8, CAF stimulates tumor growth (As is shown in the Fig. 5D).

A worse prognosis for patients has long been linked to the quantity, hardness, and other characteristics of the ECM [77]. When evaluating prognosis, it is important to measure the overall characteristics of the patient's CAFs and the prevalence of each subtype [78, 79]. Some CAF subtypes, such as iCAFs, show tumor suppressor function and are associated with improved treatment outcomes, in contrast to myCAFs and vCAFs which often suggest a poor prognosis, whereas apCAFs appear to have no prognostic implications [80, 81]. Due to the heterogeneity of CAFs and their intra-tumour specificity, some researchers have investigated their value in the early diagnosis of carcinoma and prognosis prediction as is shown in the Table 2. Cai et al. employed the Estimate the Ratio of Immune and Cancer cells (EPIC) algorithm to calculate the proportion of CAFs in patients with locally advanced rectal cancer. The results indicated a significant difference in cancer-specific survival between the two subgroups, with patients with a high rate of CAF infiltration exhibiting worse clinical outcomes [82].

The morphology of CAFs exhibits spindle-shaped, mostly polygonal, and flattened stellate forms, providing morphological cues for AI recognition of CAFs. Therefore, AI can identify CAFs and automatically quantify their numbers or ratios. Shen and associates [83] developed an imaging system which had the ability to identify CAFs with the accuracy up to 93%. By integrating with faster R-CNN cell identification technique, in the first step, extensive manual labelling of the CAF was performed on slides, and in the second step, the model was trained in conjunction with fluorescent images. The approach could significantly advance cell-based biopsies that go on to diagnose cancer. Furthermore, to extract the morphological dynamics and motility properties of cells from unlabeled live cell imaging data from CAFs, several researchers have combined a range of unsupervised and supervised machine learning methods with a deep learning-based cell categorization strategy [84]. Wu et al. [85] observed that the fibroblast growth factor receptor (FGFR) signaling pathway was enriched in the immune-exclusion phenotype of triple-negative breast cancer (TNBC) samples from the TCGA dataset after using DL to analyze the TME.

To create more potent targeted medications for immunotherapy, Charan et al. [86] used a collection of 2356 compounds to create an artificial intelligence-based prediction model for FGFR1 inhibitors. Four machine learning algorithms were used in this study, including Support Vector Machines, Random Forest, K-Nearest Neighbors, and Artificial Neural Networks. With an accuracy of 89.8%, the Random Forest model was found to be the best-performing model. In addition to this, there were other relevant studies applying AI to the detection of CAF-related genes. Lv et al. [87] discovered a unique gene signature linked to CAF that may be used as a prognostic indicator and treatment response predictor. InvasionInverse convolutional algorithms, such as the xCell algorithm, which is based on the enrichment of gene signatures, the Estimated Proportion of Immune and Cancer Cells (EPIC) algorithm, and the Microenvironmental Cell Population Counter (MCP-counter) algorithm were used to calculate the abundance of CAFs in the study.

Other cellular components

Besides studying tumour cells and immune cells in TME, AI has a little application for other cells, such as adipocytes, blood vascular endothelial cells, pericytes, neurons and nerves etc. Adipocytes provide the energy needed by cancer cells for biosynthesis through a combination of adipokine secretion, lipolysis, and reprogramming of glucose metabolism. Similarly, neurons contribute to tumorigenesis. Compared to immune cells, these other cellular components of the TME seem to be in a position of underappreciation, and as such, they are not a hotspot for AI technology research.

Limitations and prospects of artificial intelligence

Traditional TME research generally depends on laboratory techniques and animal models, but is hampered by technological limitations and resource expenses. In recent years, the rapid growth of AI has brought new concepts and approaches for investigating the tumor microenvironment. Through technologies such as ML and DL, AI can extract useful information from large amounts of data, deepen our understanding of tumours and the TME, and thus guide more precise treatment strategies and prognostic assessments [88]. The development of AI technology has brought new opportunities and challenges to oncology research. Despite the significant progress made by AI in TME research, certain obstacles still need to be overcome.

Algorithms to keep up-to-date

The complexity of the TME makes it necessary for AI algorithms to be continuously optimized and updated to adapt to changing research needs.

Data quality and standardization

Absence of standardization of data, data imbalance and heterogeneity, as well as lack of training datasets can affect model training and prediction [89]. TME research involves many types of data, such as gene expression data, immune cell infiltration data, etc., and the quality and standardization of these data may be inconsistent, which needs to be judged by experienced experts to develop uniform standards [90].

Interpretability

The black-box nature of deep learning models may limit their application in tumor microenvironment research. While these models efficiently handle complex data, understanding their decision-making processes poses challenges for clinical practitioners and researchers [91].

Generalization ability

Models perform well on TME training data for certain cancers, while may have insufficient generalization ability on new data, i.e., results on the validation set are considerably divergent from the training set results [92]. This may result in the model not performing as well as expected in real-world environments, especially when the data distribution changes.

Sample size and diversity

Building effective AI models requires large and diverse datasets. However, obtaining diverse and ample samples for tumor microenvironment research, especially for rare cancer types or specific populations, may be challenging.

Ethical and privacy concerns

Large-scale data collection and usage raise ethical and privacy issues, including data security, patient consent, and data-sharing policies. These concerns may restrict data availability and hinder research progress.

Bright future prospects

With higher efficiency and cost-effectiveness, AI reduces the subjectivity and error rate of pathologists' diagnosis, and its clinical application will gradually spread, creating an automated and intelligent diagnostic environment for us.

The combination of AI with spatial transcription and single-cell sequencing has been carried out, as in the novel self-supervised deep learning framework called BIDCell [93], which combines single-cell transcriptome data and cellular morphology information, which not only provides good segmentation of cells, but also learns to spatially discriminate between gene expression and cellular morphology, providing a direction for multidimensional characterization of TME. Combining AI with other technologies can fully utilize the advantages of each technology, which is one of the ways to apply AI in the future. Future applications of AI alone could allow for spatial analysis of all aspects of TME, replacing other costly techniques.

By identifying potential immunotherapeutic targets within the TME, discovering more biomarkers for tumors, etc., AI has the ability to be further employed in drug development [94]. In addition, fusion of multiple AI algorithms can bring significant advantages in terms of improving prediction and decision-making accuracy, reducing the risk of a single model, increasing generalization ability and transparency, etc., which is one of the prominent strategies to promote the advancement of AI applications.

Conclusions

The application of AI in the TME has now moved from macro to micro. Macroscopically, AI has been gradually applied to e.g. tumour diagnosis, tumour metastasis identification, tumour grading and staging. On the microscopic level, AI has been used to analyse the TME in a more specific and detailed way by quantifying and locating immune cells such as TILs, TANs, TAMs, etc. in the TME, as well as non-immune cells such as CAFs, etc. Therefore, this paper mainly reviews the application of AI to these four types of cells, analyzes the multidimensional characteristics of TME, as well as details the similarities and differences between single AI models and fusion AI models in the study of TME. However, the TME is very complex and still has many components that have not been covered by the study, which means that we need to further develop AI models to explore the TME in depth. Through in-depth analysis of the complexity and dynamics of the TME, AI is able to reveal the mechanisms of tumour occurrence and development and provide powerful support for tumour treatment. Nonetheless, there are still certain obstacles to be addressed. Future applications of AI in tumor microenvironment research will be increasingly comprehensive and in-depth due to the ongoing advancement of technology.

Data availability

No data was used for the research described in the article.

Abbreviations

AI:

Artificial intelligence

TME:

Tumour microenvironment

H&E:

Hematoxylin and eosin

TILs:

Tumor infiltrating lymphocytes

TAMs:

Tumor-associated macrophages

TANs:

Tumor-associated neutrophils

CAFs:

Cancer-associated fibroblasts

RNN:

Recurrent neural network

LSTM:

Long Short-Term Network

GAN:

Generative adversarial networks

FGFR:

Fibroblast growth factor receptor

MDSC:

Myeloid suppressor

IGF:

Insulin-like growth factor

PDGF:

Platelet-derived growth factor

BIG-H3:

Betaig-h3

FN:

Fibronectin

ROS:

Reactive oxygen species

MPO:

Myeloperoxidase

IFN-β:

Interferon β

MMP9:

Matrix metalloproteinase 9

VEGF:

Vascular endothelial growth factor

HGF:

Hepatocyte growth factor

BC:

Breast cancer

IHC:

Immunohistochemistry

HCC:

Hepatocellular carcinoma

HGSOCs:

High-grade serous ovarian carcinomas

OS:

Overall survival

IF:

Immunofluorescence

TSS:

Tumor-specific survival

DFS:

Disease-free survival

CRC:

Colorectal cancer

ICC:

Intrahepatic cholangiocarcinoma

PDAC:

Pancreatic ductal adenocarcinoma

TNBC:

Triple-negative breast cancer

DL:

Deep learning

CNN:

Convolutional neural network

NSCLC:

Non-small cell lung carcinoma

WSI:

Whole slide image

ML:

Machine learning

pCR:

Pathologic complete response

NAC:

Neoadjuvant chemotherapy

MIBC:

Muscle-invasive bladder cancer

SVM:

Support vector machine

kNN:

K-nearest neighbour

RF:

Random forest

References

  1. de Visser KE, Joyce JA. The evolving tumor microenvironment: from cancer initiation to metastatic outgrowth. Cancer Cell. 2023;41:374–403.

    Article  PubMed  Google Scholar 

  2. Peng H, Wu X, Liu S, He M, Xie C, Zhong R, Liu J, Tang C, Li C, Xiong S, Zheng H, He J, Lu X, Liang W. Multiplex immunofluorescence and single-cell transcriptomic profiling reveal the spatial cell interaction networks in the non-small cell lung cancer microenvironment. Clin Transl Med. 2023;13:e1155.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  3. Papalexi E, Satija R. Single-cell RNA sequencing to explore immune cell heterogeneity. Nat Rev Immunol. 2018;18:35–45.

    Article  CAS  PubMed  Google Scholar 

  4. Adan A, Alizada G, Kiraz Y, Baran Y, Nalbant A. Flow cytometry: basic principles and applications. Crit Rev Biotechnol. 2017;37:163–76.

    Article  CAS  PubMed  Google Scholar 

  5. Brown M, Wittwer C. Flow cytometry: principles and clinical applications in hematology. Clin Chem. 2000;46:1221–9.

    Article  CAS  PubMed  Google Scholar 

  6. Shaul ME, Fridlender ZG. Tumour-associated neutrophils in patients with cancer. Nat Rev Clin Oncol. 2019;16:601–20.

    Article  PubMed  Google Scholar 

  7. Rakaee M, Adib E, Ricciuti B, Sholl LM, Shi W, Alessi JV, Cortellini A, Fulgenzi C, Viola P, Pinato DJ, Hashemi S, Bahce I, Houda I, Ulas EB, Radonic T, Vayrynen JP, Richardsen E, Jamaly S, Andersen S, Donnem T, Awad MM, Kwiatkowski DJ. Association of machine learning-based assessment of tumor-infiltrating lymphocytes on standard histologic images with outcomes of immunotherapy in patients with NSCLC. JAMA Oncol. 2023;9:51–60.

    Article  PubMed  Google Scholar 

  8. Kather JN, Pearson AT, Halama N, Jager D, Krause J, Loosen SH, Marx A, Boor P, Tacke F, Neumann UP, Grabsch HI, Yoshikawa T, Brenner H, Chang-Claude J, Hoffmeister M, Trautwein C, Luedde T. Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer. Nat Med. 2019;25:1054–6.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Huang S, Yang J, Fong S, Zhao Q. Artificial intelligence in cancer diagnosis and prognosis: opportunities and challenges. Cancer Lett. 2020;471:61–71.

    Article  CAS  PubMed  Google Scholar 

  10. Arevalo J, Cruz-Roa A, Arias V, Romero E, Gonzalez FA. An unsupervised feature learning framework for basal cell carcinoma image analysis. Artif Intell Med. 2015;64:131–45.

    Article  PubMed  Google Scholar 

  11. Walsh LA, Quail DF. Decoding the tumor microenvironment with spatial technologies. Nat Immunol. 2023;24:1982–93.

    Article  CAS  PubMed  Google Scholar 

  12. Luo K, Qian Z, Jiang Y, Lv D, Zhu K, Shao J, Hu Y, Lv C, Huang Q, Gao Y, Jin S, Shang D. Characterization of the metabolic alteration-modulated tumor microenvironment mediated by TP53 mutation and hypoxia. Comput Biol Med. 2023;163:107078.

    Article  CAS  PubMed  Google Scholar 

  13. Piccialli F, Somma VD, Giampaolo F, Cuomo S, Fortino G. A survey on deep learning in medicine: why, how and when? Inform Fusion. 2021;66:111–37.

    Article  Google Scholar 

  14. Schmidhuber J. Deep learning in neural networks: an overview. Neural Netw. 2015;61:85–117.

    Article  PubMed  Google Scholar 

  15. Phan NN, Huang CC, Tseng LM, Chuang EY. Predicting breast cancer gene expression signature by applying deep convolutional neural networks from unannotated pathological images. Front Oncol. 2021;11: 769447.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  16. Janowczyk A, Madabhushi A. Deep learning for digital pathology image analysis: a comprehensive tutorial with selected use cases. J Pathol Inform. 2016;7:29.

    Article  PubMed  PubMed Central  Google Scholar 

  17. Ning F, Delhomme D, Lecun Y, Piano F, Bottou L, Barbano PE. Toward automatic phenotyping of developing embryos from videos. IEEE Trans Image Process. 2005;14:1360–71.

    Article  PubMed  Google Scholar 

  18. Bychkov D, Linder N, Turkki R, Nordling S, Kovanen PE, Verrill C, Walliander M, Lundin M, Haglund C, Lundin J. Deep learning based tissue analysis predicts outcome in colorectal cancer. Sci Rep. 2018;8:3395.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Kazeminia S, Baur C, Kuijper A, van Ginneken B, Navab N, Albarqouni S, Mukhopadhyay A. Gans for medical image analysis. Artif Intell Med. 2020;109:101938.

    Article  PubMed  Google Scholar 

  20. Xu Z, Moro CF, Bozóky B, Zhang Q. GAN-based virtual re-staining: a promising solution for whole slide image analysis. 2019; arXiv:1901.04059.

  21. Liu Y, Shi Y, Mu F, Cheng J, Chen X. Glioma segmentation-oriented multi-modal MR image fusion with adversarial learning. IEEE/CAA J Autom Sin. 2022;9:1528–31.

    Article  Google Scholar 

  22. Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Kaiser L, Polosukhin I. Attention is all you need; 2017. arXiv:1706.03762.

  23. Zhu Z, Wang Z, Qi G, Mazur N, Yang P, Liu Y. Brain tumor segmentation in MRI with multi-modality spatial information enhancement and boundary shape correction. Pattern Recogn. 2024;153:110553.

    Article  Google Scholar 

  24. Nagy ML, Juncker-Jensen A, Ovadia B, Smale R, Yamamoto K, William J, Hoe N, Padmanabhan R. Quantitative image profiling of the tumor microenvironment on double stained immunohistochemistry images using deep learning. J Clin Oncol. 2019;37:e14619.

    Article  Google Scholar 

  25. Liu Y, Ma Y, Zhu Z, Cheng J, Chen X. Transsea: hybrid CNN-transformer with semantic awareness for 3-d brain tumor segmentation. IEEE Trans Instrum Meas. 2024;73:16–31.

    Article  Google Scholar 

  26. Gao R, Yuan X, Ma Y, Wei T, Johnston L, Shao Y, Lv W, Zhu T, Zhang Y, Zheng J, Chen G, Sun J, Wang YG, Yu Z. Harnessing TME depicted by histological images to improve cancer prognosis through a deep learning system. Cell Rep Med. 2024;5:101536.

    Article  PubMed  PubMed Central  Google Scholar 

  27. Moen E, Bannon D, Kudo T, Graf W, Covert M, Van Valen D. Deep learning for cellular image analysis. Nat Methods. 2019;16:1233–46.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Manju P, Devassy BR, Rajan V, King GRG. A novel approach for nuclei segmentation using U-Net. In: 2023 International Conference on Networking and Communications (ICNWC); 2023. p. 1–6.

  29. Van Valen DA, Kudo T, Lane KM, Macklin DN, Quach NT, Defelice MM, Maayan I, Tanouchi Y, Ashley EA, Covert MW. Deep learning automates the quantitative analysis of individual cells in live-cell imaging experiments. PLoS Comput Biol. 2016;12:e1005177.

    Article  PubMed  PubMed Central  Google Scholar 

  30. Mcquin C, Goodman A, Chernyshev V, Kamentsky L, Cimini BA, Karhohs KW, Doan M, Ding L, Rafelski SM, Thirstrup D, Wiegraebe W, Singh S, Becker T, Caicedo JC, Carpenter AE. Cellprofiler 3.0: next-generation image processing for biology. PLoS Biol. 2018;16:e2005970.

    Article  PubMed  PubMed Central  Google Scholar 

  31. Zunair H, Ben HA. Sharp U-Net: depthwise convolutional network for biomedical image segmentation. Comput Biol Med. 2021;136:104699.

    Article  PubMed  Google Scholar 

  32. Stirling DR, Carpenter AE, Cimini BA. Cellprofiler analyst 3.0: accessible data exploration and machine learning for image analysis. Bioinformatics. 2021;37:3992–4.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  33. Vayrynen JP, Lau MC, Haruki K, Vayrynen SA, Dias CA, Borowsky J, Zhao M, Fujiyoshi K, Arima K, Twombly TS, Kishikawa J, Gu S, Aminmozaffari S, Shi S, Baba Y, Akimoto N, Ugai T, Da SA, Song M, Wu K, Chan AT, Nishihara R, Fuchs CS, Meyerhardt JA, Giannakis M, Ogino S, Nowak JA. Prognostic significance of immune cell populations identified by machine learning in colorectal cancer using routine hematoxylin and eosin-stained sections. Clin Cancer Res. 2020;26:4326–38.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  34. Kaneko M, Tsuji K, Masuda K, Ueno K, Henmi K, Nakagawa S, Fujita R, Suzuki K, Inoue Y, Teramukai S, Konishi E, Takamatsu T, Ukimura O. Urine cell image recognition using a deep-learning model for an automated slide evaluation system. BJU Int. 2022;130:235–43.

    Article  PubMed  Google Scholar 

  35. Li Y, Mahjoubfar A, Chen CL, Niazi KR, Pei L, Jalali B. Deep cytometry: deep learning with real-time inference in cell sorting and flow cytometry. Sci Rep. 2019;9:11088.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  36. Piansaddhayanon C, Koracharkornradt C, Laosaengpha N, Tao Q, Ingrungruanglert P, Israsena N, Chuangsuwanich E, Sriswasdi S. Label-free tumor cells classification using deep learning and high-content imaging. Sci Data. 2023;10:570.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  37. Junttila MR, de Sauvage FJ. Influence of tumour micro-environment heterogeneity on therapeutic response. Nature. 2013;501:346–54.

    Article  CAS  PubMed  Google Scholar 

  38. Brummel K, Eerkens AL, de Bruyn M, Nijman HW. Tumour-infiltrating lymphocytes: from prognosis to treatment selection. Br J Cancer. 2023;128:451–8.

    Article  CAS  PubMed  Google Scholar 

  39. Paijens ST, Vledder A, de Bruyn M, Nijman HW. Tumor-infiltrating lymphocytes in the immunotherapy era. Cell Mol Immunol. 2021;18:842–59.

    Article  CAS  PubMed  Google Scholar 

  40. van der Leun AM, Thommen DS, Schumacher TN. CD8(+) t cell states in human cancer: insights from single-cell analysis. Nat Rev Cancer. 2020;20:218–32.

    Article  PubMed  PubMed Central  Google Scholar 

  41. de Melo GD, Cortes J, Curigliano G, Loi S, Denkert C, Perez-Garcia J, Holgado E. Tumor-infiltrating lymphocytes in breast cancer and implications for clinical practice. Biochim Biophys Acta Rev Cancer. 2017;1868:527–37.

    Article  Google Scholar 

  42. Elhanani O, Ben-Uri R, Keren L. Spatial profiling technologies illuminate the tumor microenvironment. Cancer Cell. 2023;41:404–20.

    Article  CAS  PubMed  Google Scholar 

  43. Ingold HB, Untch M, Denkert C, Pfitzner BM, Lederer B, Schmitt W, Eidtmann H, Fasching PA, Tesch H, Solbach C, Rezai M, Zahm DM, Holms F, Glados M, Krabisch P, Heck E, Ober A, Lorenz P, Diebold K, Habeck JO, Loibl S. Tumor-infiltrating lymphocytes: a predictive and prognostic biomarker in neoadjuvant-treated HER2-positive breast cancer. Clin Cancer Res. 2016;22:5747–54.

    Article  Google Scholar 

  44. Lee HJ, Cho SY, Cho EY, Lim Y, Cho SI, Jung W, Song S, Kang M, Ryu J, Ma M, Park S, Paeng K, Ock C, Song SY, Gong G. Artificial intelligence (AI)-powered spatial analysis of tumor-infiltrating lymphocytes (TIL) for prediction of response to neoadjuvant chemotherapy (NAC) in triple-negative breast cancer (TNBC). J Clin Oncol. 2022;40:595.

    Article  Google Scholar 

  45. Zheng Q, Yang R, Ni X, Yang S, Jiao P, Wu J, Xiong L, Wang J, Jian J, Jiang Z, Wang L, Chen Z, Liu X. Quantitative assessment of tumor-infiltrating lymphocytes using machine learning predicts survival in muscle-invasive bladder cancer. J Clin Med. 2022;11:7081.

    Article  PubMed  PubMed Central  Google Scholar 

  46. Lu Z, Xu S, Shao W, Wu Y, Zhang J, Han Z, Feng Q, Huang K. Deep-learning-based characterization of tumor-infiltrating lymphocytes in breast cancers from histopathology images and multiomics data. JCO Clin Cancer Inform. 2020;4:480–90.

    Article  PubMed  Google Scholar 

  47. Saltz J, Gupta R, Hou L, Kurc T, Singh P, Nguyen V, Samaras D, Shroyer KR, Zhao T, Batiste R, Van Arnam J, Shmulevich I, Rao A, Lazar AJ, Sharma A, Thorsson V. Spatial organization and molecular correlation of tumor-infiltrating lymphocytes using deep learning on pathology images. Cell Rep. 2018;23:181–93.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Yanagawa N, Shikanai S, Sugai M, Koike Y, Asai Y, Tanji T, Sugimoto R, Osakabe M, Uesugi N, Saito H, Maemondo M, Sugai T. Prognostic and predictive value of CD163 expression and the CD163/cd68 expression ratio for response to adjuvant chemotherapy in patients with surgically resected lung squamous cell carcinoma. Thorac Cancer. 2023;14:1911–20.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Sousa S, Brion R, Lintunen M, Kronqvist P, Sandholm J, Monkkonen J, Kellokumpu-Lehtinen PL, Lauttia S, Tynninen O, Joensuu H, Heymann D, Maatta JA. Human breast cancer cells educate macrophages toward the M2 activation status. Breast Cancer Res. 2015;17:101.

    Article  PubMed  PubMed Central  Google Scholar 

  50. Murray PJ, Wynn TA. Protective and pathogenic functions of macrophage subsets. Nat Rev Immunol. 2011;11:723–37.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  51. Pittet MJ, Michielin O, Migliorini D. Clinical relevance of tumour-associated macrophages. Nat Rev Clin Oncol. 2022;19:402–21.

    Article  PubMed  Google Scholar 

  52. Hanna A, Metge BJ, Bailey SK, Chen D, Chandrashekar DS, Varambally S, Samant RS, Shevde LA. Inhibition of hedgehog signaling reprograms the dysfunctional immune microenvironment in breast cancer. Oncoimmunology. 2019;8:1548241.

    Article  PubMed  Google Scholar 

  53. Cassetta L, Pollard JW. A timeline of tumour-associated macrophage biology. Nat Rev Cancer. 2023;23:238–57.

    Article  CAS  PubMed  Google Scholar 

  54. Guo S, Chen X, Guo C, Wang W. Tumour-associated macrophages heterogeneity drives resistance to clinical therapy. Expert Rev Mol Med. 2022;24: e17.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  55. Yang L, Wang F, Wang L, Huang L, Wang J, Zhang B, Zhang Y. CD163+ tumor-associated macrophage is a prognostic biomarker and is associated with therapeutic effect on malignant pleural effusion of lung cancer patients. Oncotarget. 2015;6:10592–603.

    Article  PubMed  PubMed Central  Google Scholar 

  56. Khorana AA, Ryan CK, Cox C, Eberly S, Sahasrabudhe DM. Vascular endothelial growth factor, CD68, and epidermal growth factor receptor expression and survival in patients with stage II and stage III colon carcinoma: a role for the host response in prognosis. Cancer-Am Cancer Soc. 2003;97:960–8.

    Google Scholar 

  57. Ruffell B, Au A, Rugo HS, Esserman LJ, Hwang ES, Coussens LM. Leukocyte composition of human breast cancer. Proc Natl Acad Sci U S A. 2012;109:2796–801.

    Article  CAS  PubMed  Google Scholar 

  58. Li Z, Yu Q, Zhu Q, Yang X, Li Z, Fu J. Applications of machine learning in tumor-associated macrophages. Front Immunol. 2022;13:985863.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  59. Wang S, Rong R, Yang DM, Fujimoto J, Yan S, Cai L, Yang L, Luo D, Behrens C, Parra ER, Yao B, Xu L, Wang T, Zhan X, Wistuba II, Minna J, Xie Y, Xiao G. Computational staining of pathology images to study the tumor microenvironment in lung cancer. Cancer Res. 2020;80:2056–66.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  60. Rostam HM, Reynolds PM, Alexander MR, Gadegaard N, Ghaemmaghami AM. Image based machine learning for identification of macrophage subsets. Sci Rep. 2017;7:3521.

    Article  PubMed  PubMed Central  Google Scholar 

  61. Sarica A, Cerasa A, Quattrone A. Random forest algorithm for the classification of neuroimaging data in Alzheimer’s disease: a systematic review. Front Aging Neurosci. 2017;9:329.

    Article  PubMed  PubMed Central  Google Scholar 

  62. Wu XR, Peng HX, He M, Zhong R, Liu J, Wen YK, Li CC, Li JF, Xiong S, Yu T, Zheng HB, Chen YH, He JX, Liang WH, Cai XY. Macrophages-based immune-related risk score model for relapse prediction in stage I–III non-small cell lung cancer assessed by multiplex immunofluorescence. Transl Lung Cancer Res. 2022;11:523–42.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  63. Chang H, Zhu Y, Zheng J, Chen L, Lin J, Yao J. Construction of a macrophage infiltration regulatory network and related prognostic model of high-grade serous ovarian cancer. J Oncol. 2021;2021:1331031.

    Article  PubMed  PubMed Central  Google Scholar 

  64. Shen X, Wang X, Shen H, Feng M, Wu D, Yang Y, Li Y, Yang M, Ji W, Wang W, Zhang Q, Song F, Liu B, Chen K, Li X. Transcriptomic analysis identified two subtypes of brain tumor characterized by distinct immune infiltration and prognosis. Front Oncol. 2021;11:734407.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  65. Wu L, Saxena S, Awaji M, Singh RK. Tumor-associated neutrophils in cancer: going pro. Cancers (Basel). 2019;11:564.

    Article  CAS  PubMed  Google Scholar 

  66. Chen Q, Yin H, Liu S, Shoucair S, Ding N, Ji Y, Zhang J, Wang D, Kuang T, Xu X, Yu J, Wu W, Pu N, Lou W. Prognostic value of tumor-associated n1/n2 neutrophil plasticity in patients following radical resection of pancreas ductal adenocarcinoma. J Immunother Cancer. 2022;10:e005798.

    Article  PubMed  PubMed Central  Google Scholar 

  67. Arvaniti E, Claassen M. Sensitive detection of rare disease-associated cell subsets via representation learning. Nat Commun. 2017;8:14825.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  68. Yang Y, Lu C, Li L, Zheng C, Wang Y, Chen J, Sun B. Construction and multicohort validation of a colon cancer prognostic risk score system based on big data of neutrophil-associated differentially expressed genes. J Cancer. 2024;15:2866–79.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  69. Newman AM, Liu CL, Green MR, Gentles AJ, Feng W, Xu Y, Hoang CD, Diehn M, Alizadeh AA. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods. 2015;12:453–7.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  70. Huang Z, Johnson TS, Han Z, Helm B, Cao S, Zhang C, Salama P, Rizkalla M, Yu CY, Cheng J, Xiang S, Zhan X, Zhang J, Huang K. Deep learning-based cancer survival prognosis from RNA-seq data: approaches and evaluations. BMC Med Genomics. 2020;13:41.

    Article  PubMed  PubMed Central  Google Scholar 

  71. Elsayed B, Elshoeibi AM, Elhadary M, Ferih K, Elsabagh AA, Rahhal A, Abu-Tineh M, Afana MS, Abdulgayoom M, Yassin M. Applications of artificial intelligence in Philadelphia-negative myeloproliferative neoplasms. Diagnostics (Basel). 2023;13:1123.

    Article  CAS  PubMed  Google Scholar 

  72. Bi L, Gao W, Meng L, Gu G, Shi Z, Bai Y. Deep learning for discovering and identifying morphological heterogeneity of neutrophils in primary hematological diseases based on bone marrow neutrophils analysis. Bloodblood. 2020;136:18.

    Article  Google Scholar 

  73. Shao H, Gao W, Zhang Q, Li J, Zhou D, Bi L, Bai Y, Shi Z. Transfer learning for identifying morphological heterogeneity of neutrophils nuclei in hematological diseases based on nuclei semantic segmentations of bone marrow smear. Blood. 2020;136:1.

    Article  Google Scholar 

  74. Pure E, Blomberg R. Pro-tumorigenic roles of fibroblast activation protein in cancer: back to the basics. Oncogene. 2018;37:4343–57.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  75. Lavie D, Ben-Shmuel A, Erez N, Scherz-Shouval R. Cancer-associated fibroblasts in the single-cell era. Nat Cancer. 2022;3:793–807.

    Article  PubMed  PubMed Central  Google Scholar 

  76. Biffi G, Tuveson DA. Diversity and biology of cancer-associated fibroblasts. Physiol Rev. 2021;101:147–76.

    Article  CAS  PubMed  Google Scholar 

  77. Grout JA, Sirven P, Leader AM, Maskey S, Hector E, Puisieux I, Steffan F, Cheng E, Tung N, Maurin M, Vaineau R, Karpf L, Plaud M, Begue AL, Ganesh K, Mesple J, Casanova-Acebes M, Tabachnikova A, Keerthivasan S, Lansky A, Berichel JL, Walker L, Rahman AH, Gnjatic S, Girard N, Lefevre M, Damotte D, Adam J, Martin JC, Wolf A, Flores RM, Beasley MB, Pradhan R, Muller S, Marron TU, Turley SJ, Merad M, Kenigsberg E, Salmon H. Spatial positioning and matrix programs of cancer-associated fibroblasts promote t-cell exclusion in human lung tumors. Cancer Discov. 2022;12:2606–25.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  78. Geng S, Xiang T, Zhang Y, Guo P, Zhang H, Zhang Z, Gu M, Zhang K, Song H, Shi J, Liu J. Safe engineering of cancer-associated fibroblasts enhances checkpoint blockade immunotherapy. J Control Release. 2023;356:272–87.

    Article  CAS  PubMed  Google Scholar 

  79. Mao X, Xu J, Wang W, Liang C, Hua J, Liu J, Zhang B, Meng Q, Yu X, Shi S. Crosstalk between cancer-associated fibroblasts and immune cells in the tumor microenvironment: new findings and future perspectives. Mol Cancer. 2021;20:131.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  80. Ansardamavandi A, Tafazzoli-Shadpour M. The functional cross talk between cancer cells and cancer associated fibroblasts from a cancer mechanics perspective. Biochim Biophys Acta Mol Cell Res. 2021;1868:119103.

    Article  CAS  PubMed  Google Scholar 

  81. Hu B, Wu C, Mao H, Gu H, Dong H, Yan J, Qi Z, Yuan L, Dong Q, Long J. Subpopulations of cancer-associated fibroblasts link the prognosis and metabolic features of pancreatic ductal adenocarcinoma. Ann Transl Med. 2022;10:262.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  82. Cai H, Lin Y, Wu Y, Wang Y, Li S, Zhang Y, Zhuang J, Liu X, Guan G. The prognostic model and immune landscape based on cancer-associated fibroblast features for patients with locally advanced rectal cancer. Heliyon. 2024;10:e28673.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  83. Shen C, Rawal S, Brown R, Zhou H, Agarwal A, Watson MA, Cote RJ, Yang C. Automatic detection of circulating tumor cells and cancer associated fibroblasts using deep learning. Sci Rep. 2023;13:5708.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  84. Kang M, Somayadineshraj D, Min C, Shin JH. Morphodynamic and motility feature-based deep learning classification for subtypes of cancer-associated fibroblasts. Biophys J. 2023;122(3S1):145a.

    Article  Google Scholar 

  85. Wu Y, Yi Z, Li J, Wei Y, Feng R, Liu J, Huang J, Chen Y, Wang X, Sun J, Yin X, Li Y, Wan J, Zhang L, Huang J, Du H, Wang X, Li Q, Ren G, Li H. FGFR blockade boosts T cell infiltration into triple-negative breast cancer by regulating cancer-associated fibroblasts. Theranostics. 2022;12:4564–80.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  86. Charan ES, Sharma A, Sandhu H, Garg P. FGFR1Pred: an artificial intelligence-based model for predicting fibroblast growth factor receptor 1 inhibitor. Mol Divers. 2023. https://doi.org/10.1007/s11030-023-10714-7.

  87. Lv Y, Hu J, Zheng W, Shan L, Bai B, Zhu H, Dai S. A WGCNA-based cancer-associated fibroblast risk signature in colorectal cancer for prognosis and immunotherapy response. Transl Cancer Res. 2023;12:2256–75.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  88. Carini C, Seyhan AA. Tribulations and future opportunities for artificial intelligence in precision medicine. J Transl Med. 2024;22:411.

    Article  PubMed  PubMed Central  Google Scholar 

  89. Hartl D, de Luca V, Kostikova A, Laramie J, Kennedy S, Ferrero E, Siegel R, Fink M, Ahmed S, Millholland J, Schuhmacher A, Hinder M, Piali L, Roth A. Translational precision medicine: an industry perspective. J Transl Med. 2021;19:245.

    Article  PubMed  PubMed Central  Google Scholar 

  90. Brancato V, Esposito G, Coppola L, Cavaliere C, Mirabelli P, Scapicchio C, Borgheresi R, Neri E, Salvatore M, Aiello M. Standardizing digital biobanks: integrating imaging, genomic, and clinical data for precision medicine. J Transl Med. 2024;22:136.

    Article  PubMed  PubMed Central  Google Scholar 

  91. Salih A, Boscolo GI, Gkontra P, Lee AM, Lekadir K, Raisi-Estabragh Z, Petersen SE. Explainable artificial intelligence and cardiac imaging: toward more interpretable models. Circ Cardiovasc Imaging. 2023;16:e14519.

    Article  Google Scholar 

  92. Chatterjee S and Zielinski P. On the generalization mystery in deep learning; 2022. arXiv:2203.10036.

  93. Fu X, Lin Y, Lin DM, Mechtersheimer D, Wang C, Ameen F, Ghazanfar S, Patrick E, Kim J, Yang J. Bidcell: biologically-informed self-supervised learning for segmentation of subcellular spatial transcriptomics data. Nat Commun. 2024;15:509.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  94. Mak KK, Pichika MR. Artificial intelligence in drug development: present status and future prospects. Drug Discov Today. 2019;24:773–80.

    Article  PubMed  Google Scholar 

  95. Jung H, Lodhi B, Kang J. An automatic nuclei segmentation method based on deep convolutional neural networks for histopathology images. BMC Biomed Eng. 2019;1:24.

    Article  PubMed  PubMed Central  Google Scholar 

  96. Xie L, Qi J, Pan L, Wali S. Integrating deep convolutional neural networks with marker-controlled watershed for overlapping nuclei segmentation in histopathology images. Neurocomputing. 2020;376:166–79.

    Article  Google Scholar 

  97. Gurcan MN, Pan T, Shimada H, Saltz J. Image analysis for neuroblastoma classification: segmentation of cell nuclei. In: 2006 International Conference of the IEEE Engineering in Medicine and Biology Society; 2006. p. 4844–4847.

  98. Park S, Ock CY, Kim H, Pereira S, Park S, Ma M, Choi S, Kim S, Shin S, Aum BJ, Paeng K, Yoo D, Cha H, Park S, Suh KJ, Jung HA, Kim SH, Kim YJ, Sun JM, Chung JH, Ahn JS, Ahn MJ, Lee JS, Park K, Song SY, Bang YJ, Choi YL, Mok TS, Lee SH. Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as complementary biomarker for immune checkpoint inhibition in non-small-cell lung cancer. J Clin Oncol. 2022;40:1916–28.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  99. Le H, Gupta R, Hou L, Abousamra S, Fassler D, Torre-Healy L, Moffitt RA, Kurc T, Samaras D, Batiste R, Zhao T, Rao A, Van Dyke AL, Sharma A, Bremer E, Almeida JS, Saltz J. Utilizing automated breast cancer detection to identify spatial distributions of tumor-infiltrating lymphocytes in invasive breast cancer. Am J Pathol. 2020;190:1491–504.

    Article  PubMed  PubMed Central  Google Scholar 

  100. Xu H, Cha YJ, Clemenceau JR, Choi J, Lee SH, Kang J, Hwang TH. Spatial analysis of tumor-infiltrating lymphocytes in histological sections using deep learning techniques predicts survival in colorectal carcinoma. J Pathol Clin Res. 2022;8:327–39.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  101. Makhlouf S, Wahab N, Toss M, Ibrahim A, Lashen AG, Atallah NM, Ghannam S, Jahanifar M, Lu W, Graham S, Mongan NP, Bilal M, Bhalerao A, Snead D, Minhas F, Raza S, Rajpoot N, Rakha E. Evaluation of tumour infiltrating lymphocytes in luminal breast cancer using artificial intelligence. Br J Cancer. 2023;129:1747–58.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  102. Kang M, Min C, Somayadineshraj D, Shin JH. AI-driven classification of cancer-associated fibroblasts using morphodynamic and motile features. Biorxiv 2024: 2022–2024.

  103. Ali HR, Provenzano E, Dawson SJ, Blows FM, Liu B, Shah M, Earl HM, Poole CJ, Hiller L, Dunn JA, Bowden SJ, Twelves C, Bartlett JM, Mahmoud SM, Rakha E, Ellis IO, Liu S, Gao D, Nielsen TO, Pharoah PD, Caldas C. Association between CD8+ t-cell infiltration and breast cancer survival in 12,439 patients. Ann Oncol. 2014;25:1536–43.

    Article  CAS  PubMed  Google Scholar 

  104. Garnelo M, Tan A, Her Z, Yeong J, Lim CJ, Chen J, Lim KH, Weber A, Chow P, Chung A, Ooi LL, Toh HC, Heikenwalder M, Ng IO, Nardin A, Chen Q, Abastado JP, Chew V. Interaction between tumour-infiltrating b cells and t cells controls the progression of hepatocellular carcinoma. Gut. 2017;66:342–51.

    Article  CAS  PubMed  Google Scholar 

  105. Goode EL, Block MS, Kalli KR, Vierkant RA, Chen W, Fogarty ZC, Gentry-Maharaj A, Toloczko A, Hein A, Bouligny AL, Jensen A, Osorio A, Hartkopf A, Ryan A, Chudecka-Glaz A, Magliocco AM, Hartmann A, Jung AY, Gao B, Hernandez BY, Fridley BL, Mccauley BM, Kennedy CJ, Wang C, Karpinskyj C, de Sousa CB, Tiezzi DG, Wachter DL, Herpel E, Taran FA, Modugno F, Nelson G, Lubinski J, Menkiszak J, Alsop J, Lester J, Garcia-Donas J, Nation J, Hung J, Palacios J, Rothstein JH, Kelley JL, de Andrade JM, Robles-Diaz L, Intermaggio MP, Widschwendter M, Beckmann MW, Ruebner M, Jimenez-Linan M, Singh N, Oszurek O, Harnett PR, Rambau PF, Sinn P, Wagner P, Ghatage P, Sharma R, Edwards RP, Ness RB, Orsulic S, Brucker SY, Johnatty SE, Longacre TA, Ursula E, Mcguire V, Sieh W, Natanzon Y, Li Z, Whittemore AS, Anna A, Staebler A, Karlan BY, Gilks B, Bowtell DD, Hogdall E, Candido DRF, Steed H, Campbell IG, Gronwald J, Benitez J, Koziak JM, Chang-Claude J, Moysich KB, Kelemen LE, Cook LS, Goodman MT, Garcia MJ, Fasching PA, Kommoss S, Deen S, Kjaer SK, Menon U, Brenton JD, Pharoah P, Chenevix-Trench G, Huntsman DG, Winham SJ, Kobel M, Ramus SJ. Dose-response association of cd8+ tumor-infiltrating lymphocytes and survival time in high-grade serous ovarian cancer. Jama Oncol. 2017;3:e173290.

    Article  PubMed  PubMed Central  Google Scholar 

  106. Ledderose S, Rodler S, Eismann L, Ledderose G, Ledderose C. Tumor-infiltrating lymphocytes predict survival in >/= pt2 urothelial bladder cancer. Pathol Res Pract. 2022;237:154037.

    Article  CAS  PubMed  Google Scholar 

  107. Xu Y, Zeng H, Jin K, Liu Z, Zhu Y, Xu L, Wang Z, Chang Y, Xu J. Immunosuppressive tumor-associated macrophages expressing interlukin-10 conferred poor prognosis and therapeutic vulnerability in patients with muscle-invasive bladder cancer. J Immunother Cancer. 2022;10:e003416.

    Article  PubMed  PubMed Central  Google Scholar 

  108. Lee AH, Happerfield LC, Bobrow LG, Millis RR. Angiogenesis and inflammation in invasive carcinoma of the breast. J Clin Pathol. 1997;50:669–73.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  109. Feng Q, Chang W, Mao Y, He G, Zheng P, Tang W, Wei Y, Ren L, Zhu D, Ji M, Tu Y, Qin X, Xu J. Tumor-associated macrophages as prognostic and predictive biomarkers for postoperative adjuvant chemotherapy in patients with stage II colon cancer. Clin Cancer Res. 2019;25:3896–907.

    Article  CAS  PubMed  Google Scholar 

  110. Zhou SL, Zhou ZJ, Hu ZQ, Huang XW, Wang Z, Chen EB, Fan J, Cao Y, Dai Z, Zhou J. Tumor-associated neutrophils recruit macrophages and t-regulatory cells to promote progression of hepatocellular carcinoma and resistance to sorafenib. Gastroenterology. 2016;150:1646–58.

    Article  CAS  PubMed  Google Scholar 

  111. Gu FM, Gao Q, Shi GM, Zhang X, Wang J, Jiang JH, Wang XY, Shi YH, Ding ZB, Fan J, Zhou J. Intratumoral il-17(+) cells and neutrophils show strong prognostic significance in intrahepatic cholangiocarcinoma. Ann Surg Oncol. 2012;19:2506–14.

    Article  PubMed  Google Scholar 

  112. Wikberg ML, Ling A, Li X, Oberg A, Edin S, Palmqvist R. Neutrophil infiltration is a favorable prognostic factor in early stages of colon cancer. Hum Pathol. 2017;68:193–202.

    Article  CAS  PubMed  Google Scholar 

  113. Verset L, Tommelein J, Moles LX, Decaestecker C, Boterberg T, De Vlieghere E, Salmon I, Mareel M, Bracke M, De Wever O, Demetter P. Impact of neoadjuvant therapy on cancer-associated fibroblasts in rectal cancer. Radiother Oncol. 2015;116:449–54.

    Article  PubMed  Google Scholar 

  114. Kim HM, Jung WH, Koo JS. Expression of cancer-associated fibroblast related proteins in metastatic breast cancer: an immunohistochemical analysis. J Transl Med. 2015;13:222.

    Article  PubMed  PubMed Central  Google Scholar 

  115. Liu J, Chen S, Wang W, Ning BF, Chen F, Shen W, Ding J, Chen W, Xie WF, Zhang X. Cancer-associated fibroblasts promote hepatocellular carcinoma metastasis through chemokine-activated hedgehog and tgf-beta pathways. Cancer Lett. 2016;379:49–59.

    Article  CAS  PubMed  Google Scholar 

Download references

Acknowledgements

None.

Funding

This work was financially supported by the Renmin Hospital of Wuhan University Cross-Innovation Talent Project (JCRCZN-2022–015).

Author information

Authors and Affiliations

  1. Department of Pathology, Renmin Hospital of Wuhan University, 238 Jiefang-Road, Wuchang District, Wuhan, 430060, People’s Republic of China

    Ting Xie, Aoling Huang, Honglin Yan, Xianli Ju, Lingyan Xiang & Jingping Yuan

Authors
  1. Ting Xie
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Aoling Huang
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Honglin Yan
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Xianli Ju
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Lingyan Xiang
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Jingping Yuan
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

Conceptualization: T.X.; Resources: T.X., A.H.; Writing—Original Draft: T.X., A.H., H.Y.; Review of the Manuscript: X.J., L.X., J.Y.; Funding Acquisition: J.Y.. All authors have read and approved the manuscript.

Corresponding author

Correspondence to Jingping Yuan.

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors have declared that no competing interest exists.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Xie, T., Huang, A., Yan, H. et al. Artificial intelligence: illuminating the depths of the tumor microenvironment. J Transl Med 22, 799 (2024). https://doi.org/10.1186/s12967-024-05609-6

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12967-024-05609-6

Keywords

Journal of Translational Medicine

ISSN: 1479-5876

Contact us