Enhancing Testicular Ultrasound Image Classification Through Synthetic Data and Pretraining Strategies

Testicular ultrasound imaging is vital for assessing male in- fertility, with testicular inhomogeneity serving as a key biomarker. How- ever, subjective interpretation and the scarcity of publicly available data- sets pose challenges to automated classification. In this study, we explore supervised and unsupervised pretraining strategies using a ResNet-based architecture, supplemented by diffusion-based generative models to syn- thesize realistic ultrasound images. Our results demonstrate that pre- training significantly enhances classification performance compared to training from scratch, and synthetic data can effectively substitute real images in the pretraining process, alleviating data-sharing constraints. These methods offer promising advancements toward robust, clinically valuable automated analysis of male infertility. The source code is pub- licly available at https://github.com/AImageLab-zip/TesticulUS/.