Weakly-supervised learning for lung carcinoma classification using deep learning.
Fahdi Kanavati1, Gouji Toyokawa2, Seiya Momosaki3
1Medmain Research, Medmain Inc., Fukuoka, 810-0042, Japan.
Scientific Reports
|June 11, 2020
Summary
Artificial intelligence (AI) deep learning models show promise in diagnosing lung cancer. A Convolution Neural Network (CNN) accurately differentiated lung carcinoma from non-neoplastic tissue in whole slide images.
Area of Science:
- Medical Imaging
- Computational Pathology
- Artificial Intelligence in Oncology
Background:
- Lung cancer is a leading cause of cancer mortality globally.
- Accurate histopathological diagnosis is essential for effective lung cancer treatment.
- AI deep learning shows potential in medical image analysis, but validated models for pulmonary lesion diagnosis are limited.
Purpose of the Study:
- To develop and validate an AI model for the pathological diagnosis of lung carcinoma using whole slide images (WSIs).
- To assess the performance of the AI model on independent, large-scale test sets.
Main Methods:
- A Convolutional Neural Network (CNN) based on the EfficientNet-B3 architecture was trained.
- Transfer learning and weakly-supervised learning techniques were employed.
- A training dataset comprised 3,554 WSIs.
Main Results:
- The CNN model achieved high accuracy in differentiating lung carcinoma from non-neoplastic tissue.
- Excellent Receiver Operator Curve (ROC) area under the curves (AUCs) were obtained on four independent test sets (0.975, 0.974, 0.988, and 0.981).
Conclusions:
- The developed AI algorithm demonstrates significant potential for lung cancer pathological diagnosis.
- This validates the utility of AI in routine pathology, potentially reducing pathologist workload.
- Further development could lead to AI-assisted software for clinical practice.
