Interpretable Diagnosis for Whole-Slide Melanoma Histology Images Using Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Deep learning models, particularly convolutional neural networks (CNNs), show high performance in medical image diagnosis.
• The 'black-box' nature of CNNs hinders their clinical adoption and trust.
• Interpretable AI is crucial for integrating machine learning into healthcare decision-making.

Purpose of the Study:

• To propose a novel interpretable diagnosis pipeline utilizing CNNs for medical image analysis.
• To develop and validate a CNN model for accurate melanoma classification.
• To enhance clinician trust in AI diagnostic tools through model interpretability.

Main Methods:

• Development of a deep learning pipeline incorporating CNNs for medical image diagnosis.
• Creation of a large melanoma database comprising 841 digital whole-slide images (WSIs).
• Application of gradient-weighted class activation mapping (Grad-CAM) for visualizing model decision-making processes.

Main Results:

• The CNN model achieved high melanoma classification performance (AUC 0.962, sensitivity 0.887, specificity 0.925).
• The proposed model demonstrated superior accuracy (0.933) compared to 20 pathologists (0.732).
• Feature heat maps generated by Grad-CAM confirmed that the model learned clinically relevant pathological features.

Conclusions:

• The interpretable CNN model offers a rapid and accurate method for melanoma diagnosis.
• Visualizing distinctive features through saliency mapping builds confidence in AI-driven diagnostic outcomes.
• The proposed pipeline facilitates the integration of AI into clinical workflows, improving healthcare processes.