Clinician-Centric Explainable Artificial Intelligence Framework for Medical Imaging Diagnostics: A Systematic Review

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Diagnostic Support

Background:

• Medical imaging has advanced significantly, with AI, especially deep learning, playing a crucial role in diagnostics.
• Pneumonia detection is a key application area for AI in medical imaging.

Purpose of the Study:

• To systematically review AI-driven medical imaging research for pneumonia detection, focusing on classification models and explainability.
• To identify dominant AI architectures, dataset trends, performance, and challenges.
• To propose a clinician-centric explainable AI (CC-XAI) framework.

Main Methods:

• A systematic literature review (SLR) guided by PRISMA criteria.
• Synthesis of 95 studies on AI in pneumonia detection.
• Analysis of model architectures, datasets, performance metrics, and explainability approaches.

Main Results:

• Convolutional Neural Networks (CNNs) are the most common AI models used.
• High diagnostic performance (often >90% accuracy and AUC) reported for various AI models.
• Significant gaps exist in explainability, clinical integration, ethics, and trust evaluation.

Conclusions:

• Deep learning dominates AI in medical imaging for pneumonia detection.
• A clinician-centric explainable AI (CC-XAI) framework is proposed to enhance transparency and trust.
• There is an urgent need for clinician-oriented XAI to facilitate responsible AI deployment in healthcare.