A mini review examined eighteen representative studies on using Explainable AI techniques like SHAP and LIME for kidney stone segmentation. These methods integrate with deep learning and hybrid models to help identify stones in medical images. The focus was on resource-constrained settings where advanced imaging tools may be scarce. The review aimed to see if these AI tools could enhance clinician trust and support early diagnosis.
The studies looked at various AI approaches but noted several gaps. There was a lack of diversity in the datasets used, which could limit how well the tools work across different patient groups. Additionally, the research did not include multimodal integration, meaning it did not combine multiple data types for a fuller picture. Real-world validation was also scarce, as most tests occurred in controlled environments rather than busy clinics.
Safety concerns were not reported in the reviewed literature because these are software tools rather than drugs or devices. However, the absence of real-world testing means clinicians should proceed with caution. The main reason to be careful is that the current evidence is limited by data gaps and a lack of practical testing in diverse populations. Readers should view this as a promising but incomplete area of research that requires more study before widespread adoption.
