Skip to main content
Download PDF

Deep learning-based pathological prediction of lymph node metastasis for patient with renal cell carcinoma from primary whole slide images

Journal of Translational Medicine volume 22, Article number: 568 (2024) Cite this article

Abstract

Background

Metastasis renal cell carcinoma (RCC) patients have extremely high mortality rate. A predictive model for RCC micrometastasis based on pathomics could be beneficial for clinicians to make treatment decisions.

Methods

A total of 895 formalin-fixed and paraffin-embedded whole slide images (WSIs) derived from three cohorts, including Shanghai General Hospital (SGH), Clinical Proteomic Tumor Analysis Consortium (CPTAC) and Cancer Genome Atlas (TCGA) cohorts, and another 588 frozen section WSIs from TCGA dataset were involved in the study. The deep learning-based strategy for predicting lymphatic metastasis was developed based on WSIs through clustering-constrained-attention multiple-instance learning method and verified among the three cohorts. The performance of the model was further verified in frozen-pathological sections. In addition, the model was also tested the prognosis prediction of patients with RCC in multi-source patient cohorts.

Results

The AUC of the lymphatic metastasis prediction performance was 0.836, 0.865 and 0.812 in TCGA, SGH and CPTAC cohorts, respectively. The performance on frozen section WSIs was with the AUC of 0.801. Patients with high deep learning-based prediction of lymph node metastasis values showed worse prognosis.

Conclusions

In this study, we developed and verified a deep learning-based strategy for predicting lymphatic metastasis from primary RCC WSIs, which could be applied in frozen-pathological sections and act as a prognostic factor for RCC to distinguished patients with worse survival outcomes.

Background

Renal cell carcinoma (RCC) is a highly prevalent cancer, with the sixth incidence rate in male malignancies and the ninth incidence rate in female malignancies [1,2,3]. In 2021, RCC patients accounted for around 3–5% of new estimated cancer patients [1, 4]. RCC has multiple pathological subtypes, including clear cell RCC (ccRCC), papillary RCC (pRCC), chromophobe RCC (ChRCC) and other rare types [5]. Among all the histological subtypes, ccRCC is the predominant subtype and comprises up to 70–80% of all RCC cases [6]. The primary treatment of localized RCC is nephrectomy or ablation [7, 8]. Patients with localized RCC usually have acceptable clinical outcomes after treatment [9]. However, metastasis or recurrence RCC patients have extremely high mortality rate, with the 5-year overall survival around 8-11.7% [6, 10, 11]. It is estimated that around 30% of localized RCC patients eventually progress to metastasis even after treatment [7].

Lymph node involvement represents for regional spread of RCC, which may proceed to distant metastasis eventually [7, 12]. It is reported that RCC patients with lymph node involvement have worse clinical outcomes with a median recurrence-free survival of 4 months [12, 13]. However, the benefit of lymph node dissection for RCC patients remains controversial [12]. In addition, some micrometastasis in lymph nodes may be undetected by pathologists, which needs serial sections of lymph node histological slides [14]. It is necessary that an optimal criterion for lymph node dissection and serial sections of histological slides.

Intriguingly, with the further application of artificial intelligence on medical sciences, there appear some studies on prediction of lymph node involvement based on histological features [15,16,17]. These studies were targeted on prostate cancer, colorectal cancer and gastric cancer, through deep learning techniques [15,16,17]. In this study, we are the first to utilize deep learning method to predict the lymph node involvement based on whole slide image (WSI) of RCC.

Here, in this study, we developed a deep learning-based strategy for predicting lymphatic metastasis from primary WSI. We further verified the deep learning-based prediction of lymph node metastasis (DLNM) model in frozen-pathological sections and the potential clinical use of prognosis prediction of patients with RCC in multi-source patient cohorts.

Materials and methods

Data sources

Our study recruited three large patient-based cohorts from Shanghai General Hospital (SGH), Clinical Proteomic Tumor Analysis Consortium (CPTAC) [18, 19], and the Cancer Genome Atlas (TCGA) [18]. All the included patients shall have pathological diagnosis of RCC. Additional inclusion criteria for this study included: (i) with complete clinicopathological information and disease-free survival follow-up information; (ii) without severe surgical complications and other types of malignant tumors; (iii) without postoperative drug therapy (iv) with access to hematoxylin-eosin stained (H&E) slides or WSI. Basic clinical characteristics of patients from three independent patient cohorts were shown in Table S1. The TCGA cohort was randomly divided at the patient level in training set (80%) and testing set (20%) for the training and internal verification of the deep learning-based model. The SGH cohort and CPTAC cohort were used as the independent external verification cohorts.

SGH cohort

The SGH cohort recruited 486 patients who underwent partial or radical nephrectomy operative treatments and were pathologically diagnosed as RCC from January 2012 to September 2019 in SGH. After excluding the participants failed to meet the inclusion criteria, 402 cases were suitable for this study, including 307 ccRCC case, 51 pRCC cases, and 44 ChRCC cases. The corresponding H&E-stained slides of formalin-fixed and parrffin-embedded sections were retrieved from the pathology database in SGH, which were further scanned at SGH with Leica Aperio AT2 scanners at 200× equivalent magnification for the digitalization of slides.

TCGA cohort

A total of 381 patients from the TCGA database were also included in the study, which contained diagnostic pathological images met with the inclusion criteria mentioned above. The corresponding H&E-stained images with 200× equivalent magnifications were further acquired from the same database, including 307 ccRCC case, 51 pRCC cases, and 44 ChRCC cases.

CPTAC cohort

In addition, 112 cases with ccRCC from the CPTAC cohort with digitized WSIs, which were met with the inclusion criteria mentioned above, were also included. The CPTAC cohort was used as an external validation cohort to estimate the generalization performance of the DLNM model.

Image pre-processing

All available WSIs were strictly reviewed by two experienced pathologists to ensure that each slide had representative tumor regions. Since the WSIs were labeled in slide-level without manual annotations of tumor regions, we firstly segmented the WSI to remove non-tissue regions and exclude any holes. During the tessellation process, we segmented the whole slide within the segmented foreground contours into 256 × 256 patches at 100× magnification without overlap. All the patches extracted from the same slide were then identified as the instance of the WSI. Image patches and their coordinates were then stored as hdf5 hierarchical format [20].

Deep learning strategy based on multiple instance learning

In this study, we applied a clustering-constrained-attention multiple-instance learning method [20] to accurately perform instance-level clustering without any manual annotations. The overall architecture of the hybrid neural network was displayed in Fig. 1. Based on the patches of WSI, the multiple-instance learning strategy achieved the classification of WSI in view of the entire information from the slide. Firstly, we carried out dimensionality reduction from raw image data through encoding each 256 × 256 patch into a descriptive 1024-dimensional feature vector using a ResNet50-based CNN with fixed parameters pretrained on ImageNet32 [21,22,23]. Feature information from the whole region of each WSI was then assembled through attention-based pooling and contributed to predict the classification of a slide [20, 24]. Two fully-connected layers were set followed by the activation function of rectified linear unit (ReLU). Referring to the predicted attention weight of each patch, the attention pooling would average the representative features of a slide for prediction. We implied cross-entropy loss function for multiple tasks to predict tumor group.

Fig. 1
figure 1

Architecture of the deep neural network in this study

Full size image

Evaluation of the deep learning-based model

In this study, evaluation of the DLNM was performed through receiver operating characteristic curve (ROC) analysis with area under curve (AUC). The best cut-off value for the prediction model was calculated through ROC analysis, with the best specificity and sensitivity. Significant differences of AUC values were evaluated through DeLong methods [25]. Survival analysis was performed via Kaplan–Meier (KM) curve with hazard ratio (HR) and 95% confidence interval (CI) to compare different overall survival (OS) outcomes.

Results

Overall performance of the DLNM

Based on the TCGA cohort, we explored the slide-level classification performance of the deep neural network in the task of detecting lymph node metastasis (LNM) from primary tumor images. The TCGA cohort was firstly randomly partitioned into a training set (80% of cases) and a testing set (20% of cases), stratified by each class. We adopted any level of LNM as positive reference standard in the training. All models are trained for at least 50 epochs according to the monitored validation loss of each epoch [20]. The DLNM was then developed based on the optimization model with the lowest validation loss. Evaluated by the ROC analysis, our DLNM achieved an AUC of 0.836 (95% CI 0.734–0.911) in the internal testing set (Fig. 2A). When the cut-off value was set as 0.382, the DLNM achieved the best general performance with a sensitivity and specificity of 0.818 (0.482–0.977) and 0.908 (0.810–0.965), respectively.

Fig. 2
figure 2

Overall performance of the DLNM in the internal validation and external validations through receiver operating characteristic curve. (A) Evaluation of the DLNM in the TCGA testing cohort. (B) Verification of the DLNM in the SGH cohort. (C) Verification of the DLNM in the CPTAC cohort. DLNM, deep learning-based prediction of lymph node metastasis; TCGA, the Cancer Genome Atlas; Shanghai General Hospital, SGH; CPTAC, Clinical Proteomic Tumor Analysis Consortium; AUC, area under the curve with 95% confidence interval

Full size image

Validation on the external cohort

The robustness of a prediction model might be influenced by different datasets due to the data-specific variables among different patient cohorts, which arouses the importance for validating models in external patient cohorts [11]. Therefore, we recruited external validations in SGH cohort and CPTAC cohort to further evaluate the generalization performance of our DLNM. As shown in the Fig. 2B and C, the DLNM also performed well in external validations, with AUC of 0.865 (0.828–0.897) and 0.812 (0.727–0.879) in the SGH cohort and CPTAC cohort, respectively, which suggested the good generalization of the DLNM.

Application of DLNM in frozen-pathological sections

For localized RCC, surgery is the only curative treatment with high-quality evidence [26]. For patients with localized disease and clinically enlarged lymph nodes, the lymph node dissection is currently performed for staging purposes. However, as the most common method for preliminary evaluation of LNM before surgery, radiological observation is still restrained by its indirect imaging and low-resolution. Therefore, we further explored whether our DLNM could also be applied in frozen-pathological sections, which could be made in the operations and help to make auxiliary diagnosis during surgery. We retrieved another 588 WSIs from frozen RCC sections in the TCGA dataset and used the analytical framework of the same DLNM without transformation. As shown in Fig. 3, our DLNM could also be applied in frozen-pathological sections, with AUC of 0.801 (0.766–0.833), sensitivity of 0.831 (0.717–0.912) and specificity of 0.614 (0.571–0.656). Our DLNM could accurately and efficiently evaluate LNM status as well as greatly improve the surgery efficiency.

Fig. 3
figure 3

Performance of the DLNM in frozen sections of renal cell carcinoma. DLNM, deep learning-based prediction of lymph node metastasis; AUC, area under the curve with 95% confidence interval

Full size image

Prognosis prediction of patients with RCC through DLNM

Prognosis of patients with RCC is influenced by molecular and clinicopathologic factors. Among all risk factors, LNM acts as a key factor in clinical prognosis. Therefore, we further explored whether our DLNM could also provide reliable prognostic information for patients with RCC. Patients were categorized into high-DLNM or low-DLNM groups based on the best cut-off value of our model. KM curve analyses indicated that patients with high-DLNM seemed to have worse survival outcomes compared with patients with low-DLNM, with the HR of 1.84 (1.11–3.08, p = 0.0036) in the whole TCGA cohort (Fig. 4A). Validations on the external SGH cohort and the CPTAC cohort also confirmed that patients with different predicted DLNM statuses had distinct prognosis during the follow-up, with HR of 8.71 (1.62–46.90, p < 0.0001, Fig. 4A) and 5.79 (1.48–22.63, p < 0.0001, Fig. 4A), respectively. Further Cox regression analysis revealed that our DLNM could act as a prognostic factor for RCC, illustrating that our DLNM had a promising risk stratification performance in independent patient cohorts (Fig. 4B). In addition, differences in the distribution of RCC with different tumor grades between patients with high or low DLNM were also observed in the study, which revealed that our DLNM might be associated with higher levels of tumor grades (Fig. 4C).

Fig. 4
figure 4

Prognosis prediction of patients with renal cell carcinoma through the DLNM. (A) Kaplan-Meier survival analysis stratified by DLNM for overall survival in three independent patient cohorts. (B) Cox regression analysis of the DLNM for overall survival in three independent patient cohorts. (C) Differences in the distribution of renal cell carcinoma with different tumor grades between patients with high or low DLNM. DLNM, deep learning-based prediction of lymph node metastasis; TCGA, the Cancer Genome Atlas; Shanghai General Hospital, SGH; CPTAC, Clinical Proteomic Tumor Analysis Consortium; HR, hazard ratio; CI, confidence interval; ns, no significant

Full size image

Discussion

TNM classification is the most important and commonly used prognosis evaluation system [3, 27, 28]. In addition, histological features concerning prognosis should also be taken into consideration. Different pathological subtypes of RCC indicate diverse clinical outcomes [5]. Moreover, the difference in cell differentiation also results in different clinical outcomes [27, 29, 30]. Additionally, the appearance in cellular levels of RCC shows relationship with the prognosis of RCC patients [31]. The International Society of Urological Pathology (ISUP) grading system is a prognostic classification regarding the morphology of histological nuclear abnormality [2, 32, 33]. High ISUP scores indicate advanced nuclear abnormality and unfavorable prognosis [5, 32]. Clinicians make treatment decisions for RCC patients usually based on TNM classification, histological subtypes and ISUP grading [2].

With the development of artificial intelligence, deep learning techniques have been intended to be applied in the analysis of pathological images [2]. Initially researchers mainly targeted in single cell level on immunochemical images [34]. As the rapid progress in deep learning, the multicolored HE-stained whole-slide images were analyzed for various purposes, mainly for precise diagnosis, cancer prognosis prediction and drug resistance evaluation. However, limited studies were targeted on lymph node involvement prediction based on WSIs. Wang et al. evaluated the lymph node involvement of gastric cancer from lymph node WSIs [17]. Wessels et al. applied convolutional neural network to predict the lymph node involvement in prostate cancer WSIs [15]. Moreover, Brockmoeller et al. predicted the lymph node involvement in early-stage colorectal carcinoma patients WSIs through deep learning [16]. In this study, we judged the lymph node involvement of RCC through the origin RCC histological images by deep learning and verified the performance of the model in frozen-pathological section WSIs.

Lymph node involvement is a critical event for advanced RCC patients [35]. The appearance of lymph node involvement and remote metastasis represent for late stages of RCC, which usually show poor clinical outcomes and unfavored drug responses [7, 36]. Currently, the evaluation of lymph node involvement is mainly through computed tomography, magnetic resonance imaging and 18F-fluorodeoxyglucose positron emission tomography/computed tomography [7]. The golden standard of lymph node involvement assessment is histopathological confirmation based on lymph node dissection. However, it remains unclear that the role of lymph node dissection during radical RCC surgery [12]. It is estimated that only 14.8% RCC patients underwent lymph node dissection are confirmed pathological lymph node involvement [37]. The American Urological Association suggests to take lymph node dissection for RCC patients with suspicious regional lymph node involvement based on computed tomography, magnetic resonance imaging and 18F-fluorodeoxyglucose positron emission tomography/computed tomography images [12, 37]. However, there is a huge gap between clinical lymph node positive and pathological lymph node positive [13]. Furthermore, the evaluation of micrometastasis is difficult from radiological imaging [38]. Hence, an improved parameter for lymph node dissection is necessary.

Unfortunately, even pathologists still could neglect some micrometastasis in lymph nodes [14]. Considering the workload, pathological technicians usually resect one HE-slide for each lymph node sample. Nevertheless, some micrometastasis occupy very limited space of a lymph node and can be detected only after serial Sect. [14]. In addition, RCC patients with micrometastasis in lymph nodes still show poor prognosis [13]. Therefore, our DLNM could be a meaningful indicator for pathologists to apply serial sections on micrometastasis diagnosis.

According to pathological diagnosis guideline, the invasion of micro-lymphatic vessels in RCC should be mentioned in the pathological diagnosis reports [5]. Tumor cells initially invade mirco-lymphatic vessels in the kidney, transport through peripheral lymphatic vessels and eventually erode a whole lymph node [17]. As a result, the invasion of micro-lymphatic vessels can be regarded as the pre-lymph node involvement and indicated for unfavorable prognosis [39, 40]. Shoup et al. found that the presence of lymphovascular invasion showed correlation with lymph node involvement [41].

The atypical nuclear appearance, including nuclear pleomorphism, abnormal nucleus- to-cytoplasm ratio, enlarged nucleoli and pathological mitosis, is a poor prognostic indicator [41,42,43,44]. The nuclear pleomorphism stands for variance in nuclear shape and morphology [43]. Studies reveal the correlation between nuclear pleomorphism and lymph node involvement [41, 43]. In addition, Conversano et al. found that the amount of mitosis, especially pathological mitosis, was related to lymph node involvement [44].

There are some limitations in this study. Firstly, the model needs to be further evaluated among RCC patients from different hospitals and regions. Secondly, the model was based on retrospective studies. As a result, the model needs further prospective validation. Thirdly, it is also important to integrate the molecular signatures of RCC to improve the prediction accuracy of the model.

Conclusions

In conclusion, we developed and verified a deep learning-based strategy for predicting lymphatic metastasis from primary RCC WSIs, which could be applied in frozen-pathological sections and act as a prognostic factor for RCC to distinguished patients with worse survival outcomes. However, further validation of our DLNM in prospective patient cohorts should also be performed for clinical practices.

Data availability

All data generated or analyzed during this study are included in this published article.

Code availability

All code related to this method was written in Python. Custom code related to the deep learning models is available at https://github.com/mahmoodlab/CLAM.

References

  1. Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin. 2021;71(1):7–33.

    Article  PubMed  Google Scholar 

  2. Chen ST, Zhang N, Jiang LR, Gao F, Shao JL, Wang T, Zhang EC, Yu H, Wang X, Zheng JH. Clinical use of a machine learning histopathological image signature in diagnosis and survival prediction of clear cell renal cell carcinoma. Int J Cancer. 2021;148(3):780–90.

    Article  CAS  PubMed  Google Scholar 

  3. Chen S, Jiang L, Gao F, Zhang E, Wang T, Zhang N, Wang X, Zheng J. Machine learning-based pathomics signature could act as a novel prognostic marker for patients with clear cell renal cell carcinoma. Br J Cancer. 2022;126(5):771–7.

    Article  CAS  PubMed  Google Scholar 

  4. Gu J, He Z, Huang Y, Luan T, Chen Z, Wang J, Ding M. Clinicopathological and Prognostic Value of Necroptosis-Associated lncRNA Model in Patients with Kidney Renal Clear Cell Carcinoma. Dis Markers 2022, 2022:5204831.

  5. Moch H, Cubilla AL, Humphrey PA, Reuter VE, Ulbright TM. The 2016 WHO classification of Tumours of the urinary system and male genital organs-Part A: renal, Penile, and testicular tumours. Eur Urol. 2016;70(1):93–105.

    Article  PubMed  Google Scholar 

  6. Chen S, Zhang E, Jiang L, Wang T, Guo T, Gao F, Zhang N, Wang X, Zheng J. Robust prediction of prognosis and immunotherapeutic response for Clear Cell Renal Cell Carcinoma through Deep Learning Algorithm. Front Immunol. 2022;13:798471.

    Article  PubMed  PubMed Central  Google Scholar 

  7. Hsieh JJ, Purdue MP, Signoretti S, Swanton C, Albiges L, Schmidinger M, Heng DY, Larkin J, Ficarra V. Renal cell carcinoma. Nat Rev Dis Primers. 2017;3:17009.

    Article  PubMed  PubMed Central  Google Scholar 

  8. Xing XL, Liu Y, Liu J, Zhou H, Zhang H, Zuo Q, Bu P, Duan T, Zhou Y, Xiao Z. Comprehensive analysis of ferroptosis- and Immune-Related signatures to improve the prognosis and diagnosis of kidney renal clear cell carcinoma. Front Immunol. 2022;13:851312.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Tang G, Guan H, Du Z, Yuan W. Comprehensive Analysis of the butyrate-metabolism-related gene signature in Tumor Microenvironment-infiltrating Immune cells in Clear Cell Renal Cell Carcinoma. Front Cell Dev Biol. 2022;10:816024.

    Article  PubMed  PubMed Central  Google Scholar 

  10. Courtney KD, Infante JR, Lam ET, Figlin RA, Rini BI, Brugarolas J, Zojwalla NJ, Lowe AM, Wang K, Wallace EM, et al. Phase I dose-escalation trial of PT2385, a first-in-class hypoxia-inducible Factor-2alpha antagonist in patients with previously treated Advanced Clear Cell Renal Cell Carcinoma. J Clin Oncol. 2018;36(9):867–74.

    Article  CAS  PubMed  Google Scholar 

  11. Chao X, Wang P, Ma X, Li Z, Xia Y, Guo Y, Ge L, Tian L, Zheng H, Du Y, et al. Comprehensive analysis of lncRNAs as biomarkers for diagnosis, prognosis, and treatment response in clear cell renal cell carcinoma. Mol Ther Oncolytics. 2021;22:209–18.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Kaldany A, Leopold ZR, Kim JE, Patel HV, Srivastava A, Tabakin AL, Singer EA. Dissecting the role of lymphadenectomy in the management of renal cell carcinoma: past, present, and future. Kidney Cancer J. 2020;18(4):103–8.

    PubMed  PubMed Central  Google Scholar 

  13. Kuusk T, Klatte T, Zondervan P, Lagerveld B, Graafland N, Hendricksen K, Capitanio U, Minervini A, Stewart GD, Ljungberg B, et al. Outcome after resection of occult and non-occult lymph node metastases at the time of nephrectomy. World J Urol. 2021;39(9):3377–83.

    Article  PubMed  Google Scholar 

  14. Niikura H, Okamoto S, Yoshinaga K, Nagase S, Takano T, Ito K, Yaegashi N. Detection of micrometastases in the sentinel lymph nodes of patients with endometrial cancer. Gynecol Oncol. 2007;105(3):683–6.

    Article  PubMed  Google Scholar 

  15. Wessels F, Schmitt M, Krieghoff-Henning E, Jutzi T, Worst TS, Waldbillig F, Neuberger M, Maron RC, Steeg M, Gaiser T, et al. Deep learning approach to predict lymph node metastasis directly from primary tumour histology in prostate cancer. BJU Int. 2021;128(3):352–60.

    Article  CAS  PubMed  Google Scholar 

  16. Brockmoeller S, Echle A, Ghaffari Laleh N, Eiholm S, Malmstrom ML, Plato Kuhlmann T, Levic K, Grabsch HI, West NP, Saldanha OL, et al. Deep learning identifies inflamed fat as a risk factor for lymph node metastasis in early colorectal cancer. J Pathol. 2022;256(3):269–81.

    Article  CAS  PubMed  Google Scholar 

  17. Wang X, Chen Y, Gao Y, Zhang H, Guan Z, Dong Z, Zheng Y, Jiang J, Yang H, Wang L, et al. Predicting gastric cancer outcome from resected lymph node histopathology images using deep learning. Nat Commun. 2021;12(1):1637.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  18. Clark K, Vendt B, Smith K, Freymann J, Kirby J, Koppel P, Moore S, Phillips S, Maffitt D, Pringle M, et al. The Cancer Imaging Archive (TCIA): maintaining and operating a public information repository. J Digit Imaging. 2013;26(6):1045–57.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Clark DJ, Dhanasekaran SM, Petralia F, Pan J, Song X, Hu Y, da Veiga Leprevost F, Reva B, Lih TM, Chang HY, et al. Integr Proteogenomic Charact Clear Cell Ren Cell Carcinoma Cell. 2019;179(4):964–e983931.

    CAS  Google Scholar 

  20. Lu MY, Williamson DFK, Chen TY, Chen RJ, Barbieri M, Mahmood F. Data-efficient and weakly supervised computational pathology on whole-slide images. Nat Biomed Eng. 2021;5(6):555–70.

    Article  PubMed  PubMed Central  Google Scholar 

  21. He K, Zhang X, Ren S, Sun J. Deep Residual Learning for Image Recognition. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR): 2016; 2016.

  22. Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S, Huang Z, Karpathy A, Khosla A, Bernstein M. ImageNet large scale visual recognition challenge. Int J Comput Vision 2014:1–42.

  23. Lu MY, Chen TY, Williamson DFK, Zhao M, Shady M, Lipkova J, Mahmood F. AI-based pathology predicts origins for cancers of unknown primary. Nature. 2021;594(7861):106–10.

    Article  CAS  PubMed  Google Scholar 

  24. Ilse M, Tomczak JM, Welling M. Attention-based Deep Multiple Instance Learning. 2018.

  25. Sun X, Xu W. Fast implementation of DeLong’s Algorithm for comparing the areas under correlated receiver operating characteristic curves. IEEE Signal Process Lett. 2014;21(11):1389–93.

    Article  Google Scholar 

  26. Ljungberg B, Bensalah K, Canfield S, Dabestani S, Hofmann F, Hora M, Kuczyk MA, Lam T, Marconi L, Merseburger AS, et al. EAU guidelines on renal cell carcinoma: 2014 update. Eur Urol. 2015;67(5):913–24.

    Article  PubMed  Google Scholar 

  27. Beer DG, Kardia SL, Huang CC, Giordano TJ, Levin AM, Misek DE, Lin L, Chen G, Gharib TG, Thomas DG, et al. Gene-expression profiles predict survival of patients with lung adenocarcinoma. Nat Med. 2002;8(8):816–24.

    Article  CAS  PubMed  Google Scholar 

  28. Jiang LR, Zhang N, Chen ST, He J, Liu YH, Han YQ, Shi XQ, Yang JJ, Mu DY, Fu GH, et al. PD-1-Positive Tumor-Associated macrophages define poor clinical outcomes in patients with muscle invasive bladder Cancer through potential CD68/PD-1 complex interactions. Front Oncol. 2021;11:679928.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  29. Wang W, Cao K, Jin S, Zhu X, Ding J, Peng W. Differentiation of renal cell carcinoma subtypes through MRI-based radiomics analysis. Eur Radiol. 2020;30(10):5738–47.

    Article  PubMed  Google Scholar 

  30. Chen S, Jiang L, Zheng X, Shao J, Wang T, Zhang E, Gao F, Wang X, Zheng J. Clinical use of machine learning-based pathomics signature for diagnosis and survival prediction of bladder cancer. Cancer Sci. 2021;112(7):2905–14.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  31. Kim K, Zhou Q, Christie A, Stevens C, Ma Y, Onabolu O, Chintalapati S, McKenzie T, Tcheuyap VT, Woolford L, et al. Determinants of renal cell carcinoma invasion and metastatic competence. Nat Commun. 2021;12(1):5760.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  32. Yi X, Xiao Q, Zeng F, Yin H, Li Z, Qian C, Wang C, Lei G, Xu Q, Li C, et al. Computed Tomography Radiomics for Predicting Pathological Grade of Renal Cell Carcinoma. Front Oncol. 2020;10:570396.

    Article  PubMed  Google Scholar 

  33. Delahunt B, Cheville JC, Martignoni G, Humphrey PA, Magi-Galluzzi C, McKenney J, Egevad L, Algaba F, Moch H, Grignon DJ, et al. The International Society of Urological Pathology (ISUP) grading system for renal cell carcinoma and other prognostic parameters. Am J Surg Pathol. 2013;37(10):1490–504.

    Article  PubMed  Google Scholar 

  34. Pan C, Schoppe O, Parra-Damas A, Cai R, Todorov MI, Gondi G, von Neubeck B, Böğürcü-Seidel N, Seidel S, Sleiman K, et al. Deep learning reveals Cancer Metastasis and therapeutic antibody targeting in the entire body. Cell. 2019;179(7):1661–e16761619.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  35. Powles T. clinicalguidelines@esmo.org EGCEa: recent eUpdate to the ESMO Clinical Practice guidelines on renal cell carcinoma on cabozantinib and nivolumab for first-line clear cell renal cancer: renal cell carcinoma: ESMO Clinical Practice guidelines for diagnosis, treatment and follow-up. Ann Oncol. 2021;32(3):422–3.

    Article  CAS  PubMed  Google Scholar 

  36. Capitanio U, Montorsi F. Renal cancer. Lancet. 2016;387(10021):894–906.

    Article  PubMed  Google Scholar 

  37. Radadia KD, Rivera-Nunez Z, Kim S, Farber NJ, Sterling J, Falkiewicz M, Modi PK, Goyal S, Parikh R, Weiss RE, et al. Accuracy of clinical nodal staging and factors associated with receipt of lymph node dissection at the time of surgery for nonmetastatic renal cell carcinoma. Urol Oncol. 2019;37(9):577. e517-577 e525.

    Article  PubMed Central  Google Scholar 

  38. Mao X, Mei R, Yu S, Shou L, Zhang W, Li K, Qiu Z, Xie T, Sui X. Emerging technologies for the detection of Cancer Micrometastasis. Technol Cancer Res Treat. 2022;21:15330338221100355.

    Article  PubMed  PubMed Central  Google Scholar 

  39. Stacker SA, Williams SP, Karnezis T, Shayan R, Fox SB, Achen MG. Lymphangiogenesis and lymphatic vessel remodelling in cancer. Nat Rev Cancer. 2014;14(3):159–72.

    Article  CAS  PubMed  Google Scholar 

  40. Petrova TV, Koh GY. Biological functions of lymphatic vessels. Science 2020, 369(6500).

  41. Shoup M, Malinzak L, Weisenberger J, Aranha GV. Predictors of axillary lymph node metastasis in T1 breast carcinoma. Am Surg. 1999;65(8):748–52. discussion 752 – 743.

    Article  CAS  PubMed  Google Scholar 

  42. Cohen JN, Yeh I, Jordan RC, Wolsky RJ, Horvai AE, McCalmont TH, LeBoit PE. Cutaneous non-neural Granular Cell Tumors Harbor recurrent ALK gene fusions. Am J Surg Pathol. 2018;42(9):1133–42.

    Article  PubMed  Google Scholar 

  43. Lu C, Romo-Bucheli D, Wang X, Janowczyk A, Ganesan S, Gilmore H, Rimm D, Madabhushi A. Nuclear shape and orientation features from H&E images predict survival in early-stage estrogen receptor-positive breast cancers. Lab Invest. 2018;98(11):1438–48.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  44. Conversano A, Abbaci M, Karimi M, Mathieu MC, de Leeuw F, Michiels S, Laplace-Builhe C, Mazouni C. Axillary reverse mapping using near-infrared fluorescence imaging in invasive breast cancer (ARMONIC study). Eur J Surg Oncol 2022.

Download references

Acknowledgements

We appreciate the partial image data from Clinical Proteomic Tumor Analysis Consortium, the Cancer Genome Atlas, and the Cancer Imaging Archive for this study. We also appreciate for the support from the clinical postdoctoral program of Renji Hospital, Shanghai Jiao Tong University School of Medicine.

Funding

This work was supported by the National Natural Science Foundation of China (No.82002665).

Author information

Author notes

  1. Feng Gao, Liren Jiang, Tuanjie Guo, Jun Lin and Weiqing Xu equal contributors and co-first authors

Authors and Affiliations

  1. Pathology Center, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

    Feng Gao, Liren Jiang, Jun Lin, Weiqing Xu, Lin Yuan, Yaqin Han & Jiji Yang

  2. Department of Urology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

    Tuanjie Guo, Qi Pan & Xiang Wang

  3. Department of Pathology, Dongtai People’s Hospital, Dongtai, Jiangsu, China

    Enhui Chen

  4. Department of Urology, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

    Ning Zhang

  5. Department of Urology, Renji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

    Siteng Chen

Authors
  1. Feng Gao
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Liren Jiang
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Tuanjie Guo
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Jun Lin
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Weiqing Xu
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Lin Yuan
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Yaqin Han
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Jiji Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Qi Pan
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Enhui Chen
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Ning Zhang
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Siteng Chen
    View author publications

    You can also search for this author in PubMed Google Scholar

  13. Xiang Wang
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

FG and XW developed the idea for the study, contributed the central idea and provided suggestions on the project throughout the study. LJ was the leading contributor in writing the manuscript. TG was the second leading contributor for manuscript preparation and writing. JL and WX conceived the idea for the study, provided pathological suggestions and major contributors for the revision of the manuscript. LY provided pathological suggestions and reviewed WSIs. YH assisted to collect patients’ clinicopathological data. JY found out all the pathological slides of SGH. QP scanned all pathological slides of SGH. EC checked the quality and resolution of all scanned WSIs. NZ funded the research. SC conducted the deep learning analysis and was the leading contributor for revising the manuscript. All authors contributed to the revision and approved the final manuscript.

Corresponding authors

Correspondence to Ning Zhang, Siteng Chen or Xiang Wang.

Ethics declarations

Ethics approval and consent to participate

The study was conducted according to the Helsinki Declaration and was approved by the Human Ethics Committee of Shanghai General Hospital. All the participated patients were well informed and were consent to be enrolled in the study.

Consent for publication

All the authors have read the manuscript and agree to publish.

Competing interests

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Additional information

Publisher’s Note

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

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, 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 changes were made. 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/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Gao, F., Jiang, L., Guo, T. et al. Deep learning-based pathological prediction of lymph node metastasis for patient with renal cell carcinoma from primary whole slide images. J Transl Med 22, 568 (2024). https://doi.org/10.1186/s12967-024-05382-6

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12967-024-05382-6

Keywords

Journal of Translational Medicine

ISSN: 1479-5876

Contact us