Predicting Mutational Status of Driver and Suppressor Genes Directly from Histopathology With Deep Learning: A Systematic Study Across 23 Solid Tumor Types

Abstract

In the last four years, advances in Deep Learning technology have enabled the inference of selected mutational alterations directly from routine histopathology slides. In particular, recent studies have shown that genetic changes in clinically relevant driver genes are reflected in the histological phenotype of solid tumors and can be inferred by analysing routine Haematoxylin and Eosin (H&E) stained tissue sections with Deep Learning. However, these studies mostly focused on selected individual genes in selected tumor types. In addition, genetic changes in solid tumors primarily act by changing signaling pathways that regulate cell behaviour. In this study, we hypothesized that Deep Learning networks can be trained to directly predict alterations of genes and pathways across a spectrum of solid tumors. We manually outlined tumor tissue in H&E-stained tissue sections from 7,829 patients with 23 different tumor types from The Cancer Genome Atlas. We then trained convolutional neural networks in an end-to-end way to detect alterations in the most clinically relevant pathways or genes, directly from histology images. Using this automatic approach, we found that alterations in 12 out of 14 clinically relevant pathways and numerous single gene alterations appear to be detectable in tissue sections, many of which have not been reported before. Interestingly, we show that the prediction performance for single gene alterations is better than that for pathway alterations. Collectively, these data demonstrate the predictability of genetic alterations directly from routine cancer histology images and show that individual genes leave a stronger morphological signature than genetic pathways.

Self-supervised deep learning for pan-cancer mutation prediction from histopathology

“…Table 1 ). In accordance with previous studies 16 , endometrial cancer (UCEC) had the highest number of detectable mutations. N=145 out of n=321 analyzable genes had an with AUROC of over 0.60, of whom 31 had an AUROC between 0.70 and 0.80 in the external validation cohort, and an additional 14 genes had an AUROC over 0.80 ( Figure 2A ).…”

Deep learning can predict multi-omic biomarkers from routine pathology images: A systematic large-scale study

“…Compared to the other omics of the molecular landscape, a relatively lower overall predictability was found for metabolic pathways. A previous study also reported that the predictability of pathway alterations could be lower than that of altered genes, suggesting that the molecular changes at gene level are likely to be responsible for the morphological patterns that drive the predictability 29 . In our study, certain metabolic pathways that are dysregulated in tumours, such as the nuclear transport pathway and fatty acid beta-oxidation, were detectable from histology in several cancer types 50, 51 .…”

“…This may explain the reason for suboptimal results obtained with metabolomic biomarkers. On the other hand, a recent study has shown that prediction performance at the single gene level is likely to be better than that for pathway alterations 29 . This may indicate that individual genes leave a stronger morphological signature than genetic pathways in H&E-stained images.…”

“…More recently, pan-cancer studies have explored the links between genetic/molecular alterations and histomorphological features in H&E images. These studies showed that in almost all malignancies, DL methods can be used to infer a plethora of biomarkers directly from routine histology 11,14,15,[27][28][29] .…”

Developing image analysis methods for digital pathology

“…An early example demonstrated that AI techniques could predict mutations in six commonly mutated genes in lung adenocarcinoma [14]. Similar strategies have since been applied to predict mutations for many more genes across a wide range of tissue types [15][16][17][18][19]. Other studies have shown that AI may be used to directly predict patient outcome from hematoxylin and eosin [H&E] slide scans [20][21][22][23].…”