MedSightAI: An Explainable AI-Powered Chest X-Ray Diagnosis Platform That Transforms Medical AI from "Black Box" to Transparent

MedSightAI is a medical imaging diagnosis platform integrated with explainable artificial intelligence (xAI). It provides transparent diagnostic support to doctors through Grad-CAM heatmaps, concept reasoning, and knowledge graph technologies, while also building an interactive learning system for medical students. It aims to solve the "black box" dilemma of medical AI and promote the popularization of AI in clinical practice.

Main Challenges Facing Current Medical AI

• Lack of transparency: Most models cannot explain their decision-making process
• Limited trust: Doctors find it hard to verify reasoning logic
• Difficulty in interpretation: Cannot clearly mark lesion locations and severity
• Training gap: Medical students lack interactive AI-assisted diagnostic learning tools

Importance of Chest X-Ray Diagnosis

Chest X-rays are key tools for screening lung diseases such as pneumonia, tuberculosis, and pulmonary fibrosis. AI assistance without explanations will affect diagnostic efficiency and popularization speed.

Dual-Mode Design

• Doctor Support Mode: Provides similar case retrieval, automated report generation, explainable diagnostic reasoning
• Education Mode: Offers interactive exercises for medical students, real-time performance evaluation and feedback

Core Mechanisms of Explainable AI

• Grad-CAM heatmap visualization: Highlights suspected lesion areas
• Concept-based explanation: Maps features to medical concepts (e.g., pulmonary consolidation, cavity formation)
• Prototype contrast learning: Matches detection patterns with known pathological concepts

Multimodal Analysis and Knowledge Graph

Combines image analysis with clinical data (symptoms, medical history, test results), integrates the MedGemma model and knowledge graph to generate structured reports.

Image Preprocessing and Feature Extraction

Uses the DenseNet121 backbone network to perform normalization, noise removal, and contrast enhancement

Vector Retrieval and Similar Case Matching

Uses MedSigLip embedding technology to encode image features, and achieves efficient similar case retrieval through the Zilliz vector database

Feedback Loop and Continuous Learning

Doctors can adjust lesion areas and add annotations; experts update diagnostic conclusions; the system continuously improves the model through a human-in-the-loop mechanism.

• Improve diagnostic efficiency: Reduce image reading time, mark suspicious areas, lower the risk of missed or misdiagnoses
• Medical education tool: Gamified learning experience, instant scoring and personalized guidance
• Promote trust building: Transparent decision-making process and visual evidence help doctors understand the working principles of AI.

Limitations

• Currently only focuses on chest X-ray diagnosis
• Diagnostic accuracy depends on the quality and diversity of training data

Future Outlook

• Expand to other imaging modalities such as CT and MRI
• Continuously collect cases to optimize the model
• Need to strengthen privacy protection (end-to-end encryption, access control, etc.).

MedSightAI represents the shift of medical AI from pursuing accuracy to balancing explainability and human-machine collaboration. By integrating advanced technologies, it provides practical tools for doctors and creates a learning environment for medical students. Explainability will become an important standard for the value of medical AI. Only by clearly explaining diagnostic basis can AI truly become a trusted partner in clinical practice.