Deep Learning-Based Universal Expert-Level Recognizing Pathological Images of Hepatocellular Carcinoma and Beyond
Wei-Ming Chen1,2, Min Fu3, Cheng-Ju Zhang4
1Liver Research Center, The First Affiliated Hospital of Fujian Medical University, Fuzhou, China.
Frontiers in Medicine
|May 9, 2022
Summary
This study developed an artificial intelligence (AI) tool for pathological image classification. The AI demonstrated high accuracy in identifying various cancers, comparable to human experts.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Developing advanced diagnostic tools for pathological image classification is crucial.
- Transfer learning offers a promising approach for diverse tumor type analysis.
Purpose of the Study:
- To create a diagnostic tool for pathological-image classification using transfer learning.
- To apply the tool to diverse tumor types, including hepatocellular carcinoma (HCC), colorectal carcinoma, and breast invasive ductal carcinoma.
Main Methods:
- A convolutional neural network framework was trained and validated using microscopic images of liver, colorectal, and breast tissues.
- Transfer learning was employed to enhance the AI's classification performance.
- The AI framework's accuracy, sensitivity, and specificity were compared against human expert interpretations.
Main Results:
- The AI algorithm achieved high sensitivity (98.0%) and specificity (99.0%) in human-machine comparisons.
- Accuracies for classifying colorectal carcinoma and breast invasive ductal carcinoma reached 96.8% and 96.0%, respectively.
- The AI framework's performance in classifying HCC was comparable to that of human experts.
Conclusions:
- The developed AI framework shows strong potential for pathological image classification across various cancer types.
- The AI's performance is comparable to human experts, suggesting its utility in clinical diagnostics.
- Future research can explore extending this AI for diagnoses and treatment recommendations.
