Grad-CAM-Based Explainable Artificial Intelligence Related to Medical Text Processing.

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Computer Vision

Background:

• Deep learning's opacity hinders medical applications, necessitating explainable AI (XAI).
• XAI ensures human understanding of AI model decisions in healthcare.
• Visualizing model attention is crucial for trust and validation.

Purpose of the Study:

• To develop and evaluate an XAI system for medical text classification.
• To enhance the transparency of deep learning models in healthcare.
• To intuitively present the basis for AI-driven medical text analysis.

Main Methods:

• Transfer learning from computer vision (ResNet) to medical text tasks.
• Utilized gradient-weighted class activation mapping (Grad-CAM) for visualization.
• Compared Word2Vec, BERT, ResNet, 1D CNN, and Bi-LSTM for text classification.

Main Results:

• Pre-trained ResNet on formalized medical text achieved the highest performance (90.9% recall, 91.1% precision, 90.2% F1-score).
• Grad-CAM visualization effectively highlighted important words in model predictions.
• The developed system demonstrated high-accuracy classification with interpretable results.

Conclusions:

• ResNet combined with Grad-CAM offers a robust XAI solution for medical text classification.
• The approach improves the interpretability and trustworthiness of AI in medicine.
• This method provides intuitive insights into AI decision-making processes for medical applications.