From: Pathogenomics for accurate diagnosis, treatment, prognosis of oncology: a cutting edge overview
Tumor type | Models | References | Data modalities | Performance |
---|---|---|---|---|
HGSOC | ResNet-18, Cox | Kevin et al. [41] | WSI, CT, CE-CT, and NGS | Survival prediction, C-index: 0.61 |
BRCA and LUAD | Inception v3 | Alona et al. [42] | WSI, RNA, and miRNA, | miR-17 status prediction, AUC: 0.95 |
Pan-cancer | Shufflenet, densenet, inception, and resnet | Jakob et al. [14] | WSI and CNV | MSI status prediction, AUC: 0.89 |
LUAD | ResNet18, GO, and KEGG | Yi et al. [53] | WSI, mRNA, and clinical characteristics | TMB status prediction, AUC: 0.64 |
Pan-cancer | HE2RNA | Benoît et al. [39] | WSI and RNA-Seq | The model could predict RNA-Seq profiles from WSIs |
STAD | ResNet18, AlexNet, vgg19, squeezenet | Jakob et al. [38] | WSI, MSI and clinical information | Shown that deep residual learning can predict MSI directly from H&E histology |
LUAD and LUSC | Inception v3 | Nicolas et al. [33] | WSI and gene mutation data | 3-Class classification performance reached AUC: 0.97; gene mutations AUCs from 0.733 to 0.856 |
Pan-cancer | Deep transfer learning, inception V4, COX | Yu et al. [13] | WSI, genomics, transcriptomics and survival data | Shown that deep learning could characterize the molecular bias of tumor pathology |
COAD and READ | GCN | Ding et al. [54] | WSI and gene mutation data | Shown that GCN models could predict molecular profile from WSIs |
KIRC | LR, RF, SVM, Adaboost | Zheng et al. [55] | WSI and DNA methylation | Shown that ML algorithms can associate the DNA methylation with histological features |