Related Experiment Video
Updated: May 22, 2026
09:16
Supervised Machine Learning for Semi-Quantification of Extracellular DNA in Glomerulonephritis
Published on: June 18, 2020
GDT-SwinKid: A hybrid model for precise renal lesion analysis.
Thirupathi Rao N1, V V Ramana Ch2, Eatedal Alabdulkreem3
1Department of Computer Science and Engineering, Vignan's Institute of Information Technology (A), Visakhapatnam, Andhra Pradesh, India.
Plos One
|May 20, 2026
Summary
A new AI model, GDT-SwinKid, enhances kidney lesion detection and classification using advanced statistical methods and transformer networks. This tool improves diagnostic accuracy for renal lesions, offering a new standard in kidney imaging analysis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biostatistics
Background:
- Accurate detection and delineation of renal lesions are challenging due to diverse kidney pathologies and subtle CT image variations.
- Existing methods struggle with precise feature extraction and contextual awareness for kidney lesion analysis.
Purpose of the Study:
- To present GDT-SwinKid, a novel hybrid AI model for improved renal lesion segmentation and classification.
- To address limitations in current kidney imaging analysis through advanced feature extraction and contextual understanding.
Main Methods:
- Integration of Swin Transformer's hierarchical attention with a modified U-Net decoder.
- Application of adaptive Gamma distribution for statistical modeling and feature refinement.
- Utilizing cross-attention mechanisms for enhanced contextual awareness.
Main Results:
- GDT-SwinKid achieved high performance with Dice coefficients up to 0.95 and AUC values near 0.99.
- Demonstrated significant improvement (5-9% Dice coefficient) over conventional U-Net and Swin Transformer baselines.
- Outperformed existing hybrid and transformer-based methods on clinical CT kidney datasets.
Conclusions:
- GDT-SwinKid offers a new standard for automated kidney lesion analysis, combining statistical sensitivity and hierarchical attention.
- Explainable attention maps and deep supervision enhance trust and facilitate integration into diagnostic workflows.
- The model increases the reliability and utility of AI techniques in renal imaging.
