AI-Driven rTMS Target Localization and Brain Segmentation via Deep Learning

This repository presents a production-ready, deep learning-driven pipeline optimized for automatic brain segmentation and target coil localization in Repetitive Transcranial Magnetic Stimulation (rTMS) therapy. Leveraging a customized U-Net encoder-decoder topology with skip-connections, the architecture counters severe class spatial imbalances through a specialized Dice Loss optimization tracker. To optimize resource allocation and circumvent local storage constraints, the data engine utilizes an advanced IterableDataset cloud-streaming pipeline to dynamically fetch and process genuine 2D human brain MRI matrices directly from cloud repositories in real-time. Upon segmenting target anatomical boundaries, the system applies a multi-panel visualization interface and high-fidelity center-of-mass spatial extraction logic to map exact targeting coordinates (\(X\) and \(Y\)). The resulting diagnostics dashboard tracks model error optimization curves against true references, providing an end-to-end framework for precision neuroanatomical targeting.