Enhanced Testicular Torsion Detection Using Discrete Wavelet Transform and Machine Learning: A Case Study From Jordan

Abstract

Testicular Torsion (TT) is a medical emergency in urology caused by the twisting of the spermatic cord leading to restricted blood flow to the testicle. TT can result in irreversible testicular damage or loss if not promptly diagnosed and surgically treated. Early and accurate diagnosis is therefore essential to preserve testicular viability. While medical image analysis has proved valuable in diagnostics, the automated detection of TT remains a challenge due to subtle image variations and limited datasets. In this study, we present a dataset of 323 ultrasound images collected from three hospitals across Jordan, encompassing both normal and TT cases. We evaluated five classification models, including two traditional machine learning models employing a novel Discrete Wavelet Transform (DWT)-based feature extraction method inspired by EEG signal analysis, and three transfer learning models based on pretrained deep architectures. Among the evaluated models, Support Vector Machine (SVM) with DWT and ResNet50 achieved the best performance, with accuracies of 88% and 87% and F1-scores of 82% for both. The results demonstrate that both the wavelet-based SVM model and the deep transfer learning model (ResNet50) achieved comparable and reliable performance for automated TT detection.