Researchers unveil XEns-CKD for kidney disease detection
AI model XEns-CKD detects kidney disease stages with 86.36% accuracy...
A team of researchers from multiple universities has developed XEns-CKD, a novel explainable ensemble-based approach for detecting chronic kidney disease (CKD) stages using ultrasound images. The model leverages three vision transformers (ViTs) trained on a private ultrasound dataset, achieving an overall classification accuracy of 86.36% across five CKD stages and normal kidney status. This represents a 4% improvement over existing methods, addressing a critical gap in early CKD detection.
The model incorporates explainable AI techniques—LIME, LRP, Attention-Min, and Attention-Max—to improve transparency and clinical trust. An attention map combining Attention-Min and Attention-Max results highlights kidney regions affected by CKD progression, enabling better interpretation of disease severity. The approach emphasizes early detection, which is crucial for slowing CKD progression and guiding patient care.
- XEns-CKD achieves 86.36% accuracy in CKD stage detection, outperforming prior methods by 4%.
- Uses explainable AI (LIME, Attention-Min/Max) to highlight affected kidney regions, improving clinical trust.
- Three ViTs trained on a private ultrasound dataset; ensemble model enhances classification performance.
Why It Matters
Early CKD detection via AI could revolutionize patient care by enabling timely intervention and slowing disease progression.