Mini review of Explainable AI for kidney stone segmentation in resource-constrained settings

Photo by CDC / Unsplash

Frontiers in Medicine Published May 17, 2026 Medically reviewed May 17, 2026 Study authors: Md Jakir Hossen, B. M. Taslimul Haque, Hasanul Bannah, Md Arifur Rahman, Abir Ahmed, Abdullah Al Nom… DOI ↗ By Dr. Amelia Tan, PhD · Internal Medicine & Chronic Disease

Key Takeaway

Note that XAI for kidney stone segmentation faces dataset diversity and validation limitations.

This mini review examines the application of Explainable AI techniques, including SHAP, LIME, Grad-CAM, Layer-wise Relevance Propagation, and EigenCAM, within deep learning and hybrid models for kidney stone segmentation. The scope focuses on resource-constrained settings where such technologies aim to improve diagnostic workflows. The authors note that these approaches are intended to enhance clinicians' trust and support clinical decision-making.

The review identifies several key limitations that constrain current implementation. These include a lack of diversity in datasets, an absence of multimodal integration, and a scarcity of real-world validation. Because these gaps exist, the synthesized evidence does not yet support broad clinical adoption without further verification.

The authors conclude that while these AI tools offer potential benefits for early diagnosis, their utility remains theoretical in many contexts. Practice relevance is currently limited to enhancing trust rather than replacing standard diagnostic protocols. Clinicians should await more robust validation before integrating these specific XAI methods into routine care.

Study Details

Study typeSystematic review

EvidenceLevel 1

PublishedMay 2026

View Original Abstract ↓

Kidney stones are one of the most common renal disorders that can produce severe complications if not diagnosed and treated early. Recently, advances in AI have ensured that deep learning and explainable AI enable the automatic segmentation and detection of kidney stones from medical imaging, thus improving diagnostic efficiency and accuracy. For this review, eighteen representative studies using machine learning, deep learning, and hybrid models for kidney stone segmentation were considered, which were published in the period between 2020 and 2025. The XAI techniques being mainly utilized with the discussed models in the study are SHAP, LIME, Grad-CAM, Layer-wise Relevance Propagation, and EigenCAM. Such approaches tend to enhance clinicians’ trust in allowing early diagnosis and supporting clinical decision-making, especially in resource-constrained settings. Regardless of the towering results, this area still suffers due to certain limitations such as lack of diversity in datasets, absence of multimodal integration, and scarcity of real-world validation. All in all, integrating DL with XAI presents a transparent, reliable, and clinically acceptable approach to detecting and segmenting kidney stones.