Ultrasound-Based Kidney Stone Classification Using Kronecker Self-Organizing Map Forward Harmonic Network.
Authors
Affiliations (5)
Affiliations (5)
- Department of CSE-AI & DS, GMR Institute of Technology, Rajam, Andhra Pradesh, India.
- Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, Karnataka, India.
- Department of Computer Science and Engineering, Anil Neerukonda Institute of Technology and Sciences, Visakhapatnam, Andhra Pradesh, India.
- Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Guntur, Andhra Pradesh, India.
- Department of CSE, GITAM School of Computer Science & Engineering, GITAM (Deemed to be University), Visakhapatnam, Andhra Pradesh, India.
Abstract
Kidney stone disease is a prevalent urological disorder that can result in severe pain, obstruction, and long-term complications if not detected and managed promptly. Traditional diagnostic approaches, particularly those relying on manual assessment of ultrasound images, often suffer from limitations such as subjective interpretation, dependency on radiologist expertise, and challenges in identifying small or complex stones. These constraints can lead to diagnostic delays and inconsistencies, especially in time-sensitive or resource-limited clinical settings. Therefore, the need for an intelligent, automated solution that enhances diagnostic accuracy and efficiency is more critical than ever. To address these issues, we propose a novel deep learning-based model called the Kronecker Self-Organizing Map Forward Harmonic Network (KSOMFHNet) for kidney stone classification using ultrasound imagery. The model begins with an image preprocessing phase, where a double bilateral filter is applied to effectively denoise the ultrasound images. Following this, the Deep Recursive Residual Network (DRRN) is employed to segment the kidney region accurately. Feature extraction is then performed using a combination of Binary Robust Independent Elementary Features (BRIEF), shape-based features, and Gray Level Co-Occurrence Matrix (GLCM) texture descriptors. These features are then used for classification via the KSOMFHNet, a hybrid architecture integrating the Deep Kronecker Neural Network (DKN) and Self-Organizing Map Network (SOMNet). This fusion enhances the model's learning capacity and spatial representation abilities. Experimental results demonstrate that KSOMFHNet achieves high performance, with an accuracy of 91.984%, a True Positive Rate (TPR) of 90.543%, a True Negative Rate (TNR) of 92.248%, a precision of 90.179%, and an <i>F</i>1-score of 90.360% for training data is 90%, highlighting its potential for clinical deployment.