A Systematic Review on Adversarial Attack Detection in Medical Image Classification

Currently, deep learning models are becoming more common in medical image classification. This has raised several concerns with respect to attack and manipulation, which may result in misdiagnosis and the endangerment of patient safety. This research proposes an explainable and robust deep learning framework for attack detection in medical imaging systems. The framework fuses advanced feature analysis, anomaly detection, and interpretability techniques to detect adversarial inputs and link them with clear explanations about the model decisions. While effectively detecting attacks through the proposed robustness measures and explainable AI approaches, it also strengthens clinician trust in automated diagnostic tools. Experimental evaluations on benchmark medical imaging datasets show that the framework detects adversarial attacks accurately while maintaining classification performance, indicating its readiness for deployment into real-world secure and reliable healthcare applications. To address these challenges, recent research has focused on combining robust deep learning techniques with explainable artificial intelligence (XAI) methods to improve both security and transparency. This review represents a comprehensive analysis of explainable and robust deep learning frameworks for adversarial attack detection in medical image classification. The paper systematically examines adversarial threat models, robustness-enhancing defense strategies, and widely used explanation techniques in medical imaging. A comparative literature study highlights existing approaches, their strengths, limitations, and research gaps.