Kidney CT scans provide valuable information for identifying conditions such as normal kidneys, cysts, tumors, and stones. Accurate interpretation of medical images, however, requires careful analysis and medical expertise.
Naveed Sultan from the Computer Engineering Department at King Mongkut’s University of Technology Thonburi (KMUTT) and the research team developed RenalRQKV-Net, a framework that combines three image encoders: ConvNeXt-Base, ConvNeXt-Hybrid, and DINOv3 ViT-B/14. The framework analyzes images from multiple perspectives and integrates their representations through a mechanism called the Receptive QKV Refiner.
The framework achieved 99.63% classification accuracy on the CTKidney dataset. The research paper also received the Best Paper Award at EECSI 2026 in Indonesia.
This achievement highlights an approach to medical AI that combines complementary model perspectives to support medical image analysis.