Classification and mutation prediction based on histopathology H&E images in liver cancer using deep learning
Mingyu Chen1,2,3, Bin Zhang1, Win Topatana4
1Department of General Surgery, Sir Run-Run Shaw Hospital, Zhejiang University, 310016 Hangzhou, China.
NPJ Precision Oncology
|June 19, 2020
Summary
This study developed an AI model for liver cancer (hepatocellular carcinoma) diagnosis using histopathology images. The AI achieved high accuracy in classifying tumors and predicting gene mutations, aiding pathologists.
Area of Science:
- Oncology
- Computational Biology
- Digital Pathology
Background:
- Hepatocellular carcinoma (HCC) diagnosis relies on pathologist expertise for grading.
- Accurate histopathological assessment is crucial for effective HCC treatment and prognosis.
Purpose of the Study:
- To develop and evaluate a deep learning model for automated HCC classification and gene mutation prediction.
- To assess the model's performance against experienced pathologists.
Main Methods:
- Utilized histopathological H&E images from the Genomic Data Commons Databases.
- Trained an Inception V3 neural network for image classification.
- Developed a model to predict common and prognostic mutated genes in HCC.
Main Results:
- Achieved 96.0% accuracy in benign vs. malignant classification and 89.6% for tumor differentiation.
- Model performance approximated that of a pathologist with 5 years of experience.
- Successfully predicted four key HCC mutated genes (CTNNB1, FMN2, TP53, ZFX4) from images with AUCs of 0.71–0.89.
Conclusions:
- Convolutional neural networks can effectively assist pathologists in HCC classification.
- AI models show potential for predicting gene mutations directly from histopathology images.
- This technology can enhance diagnostic accuracy and efficiency in liver cancer management.
