KIDNEY STONE DETECTION USING A NOVEL SHUFFLEV2 RESIDUAL NETWORK ON ULTRASONIC IMAGE

Kidney stones are calcified mineral deposits in the kidneys that can produce acute pain, blood in urine, and long-term kidney impairment if not detected early. Detecting these stones accurately from ultrasound images is often challenging, as conventional methods struggle with noise, overlapping structures, and variations in stone appearance. To address these challenges, a ShuffleV2 Residual Network (ShV2R-Net) is explored in this work for automated kidney stone detection from ultrasound images. Here, an ultrasound kidney image is composed and passed to image denoising for eliminating unwanted noise using anisotropic diffusion. Then, kidney region segmentation using Multi- Convolutional Channel Residual Spatial Attention U-Network with focal loss and trained based on a Giant Trevally Optimizer (MCRSAU-Net_Fl-GTO) isolates the kidney from surrounding structures in the image. Besides, features, like gray-level co-occurrence matrix (GLCM) texture features, discrete wavelet transform, and shape features are extracted. Finally, kidney stones are detected using ShV2R-Net, which is developed by fusing Shuffle Network Version 2 (ShuffleNet V2), deep residual network, and harmonic analysis. Moreover, ShV2R-Net acquired a maximum true positive rate of 95.654%, accuracy of 93.765%, and true negative rate of 91.876% with K-fold 8. The proposed approach offers a practical and efficient tool for clinicians, enabling early diagnosis, reducing diagnostic errors, and supporting timely treatment planning for patients with kidney stones.