Low-Cost Handheld Ultrasound for Rural Healthcare: Sonolite

Access to prenatal ultrasound diagnostics remains limited in many rural areas, primarily due to the prohibitive cost of imaging systems and the lack of specialized personnel. This paper proposes Sonolite, a lightweight, AI-enabled software framework designed to facilitate fetal structure detection from ultrasound images, with a focus on deployment in low-resource environments. The system utilizes a Faster Region-Based Convolutional Neural Network (R-CNN) model with a custom ResNet-34 backbone, optimized to detect key fetal anatomical regions, including the head, abdomen, hands, and legs. The model was trained and evaluated on a curated ultrasound dataset, achieving an average accuracy over 90%. The software is designed for real-time inference and has been integrated with a mobile interface using FastAPI and Retrofit to ensure platform compatibility. A preliminary survey of rural public health centers was conducted to identify infrastructural constraints, which guided the design considerations of the framework. Future work will focus on integrating the system with handheld ultrasound devices and conducting in-field validation to assess usability and performance in practical clinical settings.