Authors
Department of Information Technology Management, Technical College of Management/Mosul, Northern Technical University, Mosul, 41001, Iraq
[email protected]
Abstract
Kidney disease diagnosis often relies on expert radiologists and advanced imaging analysis, which may be limited or unavailable in remote healthcare settings. This paper presents an FPGA-oriented computer-aided diagnosis scheme for binary kidney abnormality classification (normal vs. abnormal) that combines kidney ROI extraction with nonsubsampled contourlet transform (NSCT) feature compaction and a modified YOLOv11 classifier. After normalization and kidney segmentation, NSCT is applied to the ROI and only the final low-frequency sub-band is retained as a compact, structure-preserving representation for classification. The key novelty is a lightweight YOLOv11 classification variant tailored for deployment constraints by reducing depth scaling and removing attention modules to support efficient fixed-point implementation while retaining discriminative power with NSCT low-frequency inputs. The proposed method has been achieved better results as shown in accuracy, sensitivity, and specificity (i.e.: 97.45%, 98.23%, and 97.45%) respectively. The higher performance of Zynq based hardware shows 68 ms latency in image, and 14.7 images/ throughput, performing better real-time for kidney CT disease classification.