DeLTA-BIT: a deep-learning based on Local TrActography for BraIn Targeting in the treatment of movement disorders.

Essential tremor (ET) is the most common adult movement disorder, markedly impairing quality of life. For medication-refractory tremor, surgical targeting of the thalamus ventral intermediate nucleus (VIM) is an effective treatment strategy. Transcranial MRI-guided focused ultrasound (tcMRgFUS) offers a non-invasive alternative, but accurate VIM localization remains challenging due to its small size and anatomical variability. Probabilistic tractography enables patient-specific targeting but is computationally demanding and impractical for real-time use. We developed DeLTA-BIT, an open-source framework for rapid VIM identification, based on a 3D UNet pre-trained on T1-w images and tractography-derived VIM labels from the Human Connectome Project, then fine-tuned on clinical tcMRgFUS datasets from ET patients. Performance was assessed on a clinical dataset (manually contoured) and compared with an atlas-based method (THOMAS). Fine-tuning on clinical data improved VIM localization accuracy over both the pre-trained model and THOMAS (DSC: 0.43±0.17, 0.23±0.11 and 0.31±0.07, respectively). VIM predictions are generated in 1 min, making DeLTA-BIT a promising tool for real-time targeting.