A Deep Diagnostic Framework Using Explainable Artificial Intelligence and Clustering.

Area of Science:

• Medical imaging analysis
• Computational pathology
• Artificial intelligence in diagnostics

Background:

• Diagnosing diseases requires understanding characteristic properties, which is challenging with image data.
• Current methods rely on manually extracted features, limiting new disease insights.
• A need exists for automated approaches to analyze complex medical images.

Purpose of the Study:

• To propose a novel framework for disease insight discovery from medical images.
• To overcome limitations of handcrafted features and human intervention in image analysis.
• To develop an automated method for differentiating healthy from pathological images.

Main Methods:

• Utilized deep learning (DL) for pattern recognition in medical images.
• Employed explainable artificial intelligence (XAI) for pattern visualization.
• Introduced a novel "explanation-weighted" clustering technique for patient data overview.

Main Results:

• The framework successfully differentiated between healthy and pathological gastrointestinal images.
• The method organized images based on the specific reasons for pathological diagnosis.
• Achieved high cluster quality and a Rand index close to one, indicating effective organization.

Conclusions:

• The proposed framework offers a powerful, automated approach to gain insights from medical images.
• Deep learning, XAI, and clustering can be combined to advance disease characterization.
• This method has significant potential for improving diagnostic accuracy and discovering novel disease properties.