Multi-feature Response Attention Network for kidney stone detection and measurement of stone size in ultrasound images

Kidney stones are hard deposits formed in the urinary tract due to excess waste and low fluid intake, causing severe abdominal pain. Ultrasound-based kidney stone detection involves identifying and locating stones in the kidneys using high-frequency sound waves to visualize renal structures. Measuring stone size is crucial for early treatment and preventing infection. However, detection is challenging because ultrasound images suffer from noise, lower contrast, and variable quality. To overcome these limitations, the Multi-feature Response Attention Network (MFRANet) is proposed to detect kidney stones. Initially, the input image is allowed for image denoising and edge preservation using Windowed Variation Kernel Wiener Filter (WV-KWF). Further, the kidney region is segmented by the Fuzzy Clustering algorithm with Variable Multi-pixel Fitting spatial information (FCM-VMF). Moreover, descriptor computation is processed by rotated Binary Robust Independent Elementary Feature (rBRIEF) with Discrete Cosine Transform (DCT). Afterwards, the kidney stone is detected using MFRANet, wherein MFRANet is designed by combining Multi-feature Multi-Attention Network (MFMANet), along with Group Response Attention Network (GRA-Net). Next, the detected stone is segmented using Optimized DA-TransUNet (O-DAT), and lastly, the kidney stone size is measured. Performance of the proposed MFRANet is assessed with Accuracy, F1-score, Precision, and True Positive Rate (TPR) which achieved superior values of 96.64%, 96.65%, 97.44%, 95.88% Response Attention Network, Multi-feature Multi-Attention Network, Optimized DA-TransUNet.