Classification and mutation prediction based on histopathology H&E images in liver cancer using deep learning
Problem
Grading hepatocellular carcinoma requires visual inspection by experienced pathologists, and mutated gene status normally requires genomic testing.
Why it matters
Accurate grading and mutation information guide liver cancer treatment, and pathologist expertise is limited.
What the AI did
Trained an Inception V3 neural network on H&E histopathology images to classify HCC and predict the ten most common prognostic mutated genes.
Results
96.0% accuracy benign vs malignant; 89.6% accuracy for tumor differentiation; external AUCs 0.71-0.89 for CTNNB1, FMN2, TP53, ZFX4; performance near a 5-year-experience pathologist by Matthews correlation coefficient.
Impact
Shows CNNs can assist pathologists with grading and mutation prediction from routine H&E slides.
Record
| Venue | npj Precision Oncology |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Mingyu Chen, Bin Zhang, Win Topatana, Jiasheng Cao, He-pan Zhu, Sarun Juengpanich, Qijiang Mao, Hong Yu et al. |
| Citations | 244 (OpenAlex, at import time) |
| References | 34 |
| Model or method | Inception V3 |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41698-020-0120-3 |
| Open access | gold |
| Keywords | liver cancer, histopathology, deep learning, mutation prediction, Inception V3 |
Importance rating
significant — thousands to millions affected
HCC described as most common liver cancer subtype; grading and mutation detection guide treatment, but abstract states no patient numbers or scale.