Fada: Fetal Accurate Detection AI for Automated Ultrasound Image Analysis and Reporting

This study introduces Fetal Accurate Detection AI (FADA) an advanced AI-driven framework for generating clinically relevant descriptions from fetal ultrasound images, specifically focused on diverse anatomical structures and views, including trans-abdominal and trans-vaginal imaging modalities. Leveraging the Bootstrapping Language-Image Pre-training (BLIP) architecture, we fine-tuned the model on a dataset comprising 38,723 images. Our methodology incorporates Low-Rank Adaptation (LoRA) for efficient parameter tuning in the image encoder, paired with specialized enhancements to the text decoder for medical vocabulary integration and structured reporting. The system demonstrated extremely high performance, achieving BLEU scores of 0.984 on the validation set and 0.9589 on the test set, showcasing its efficacy in aligning generated descriptions with expert annotations. This innovation bridges the gap between computational advancements and clinical utility, setting a new benchmark for automated medical reporting in prenatal care.