A Novel Approach for Personalized DLPFC Localization in Neuroimaging and Brain Stimulation Using Mask R-CNN

The exact location of the left dorsolateral prefrontal cortex (DLPFC) is important for proper brain stimulation and neuroimaging in neuropsychiatric work. Despite its significance, individual anatomical differences makes targeting the DLPFC difficult with conventional methods as the “ 5 cm rule” which often misses the true DLPFC location. For this reason, we propose a novel application of machine learning; specifically the Mask R-CNN deep learning architecture to automatically identify and segment the DLPFC on structural brain images. We gathered annotated brain images by experts with DLPFC labels and fine-tuned a Mask R-CNN model to detect the anatomical patterns that defines the DLPFC in each person. The trained model produces a segmented mask of the DLPFC for each image, enabling personalized targeting. Results: The model achieved high accuracy in localizing the DLPFC, with a mean average precision (mAP) of $\sim 0.85$ on validation scans, indicating close correspondence between the predicted masks and expert annotations. Significance: This automated approach outperforms heuristic methods in precision and consistency, and it offers a fast, reproducible means of tailoring targeting to each patient's brain anatomy. We discuss technical challenges such as data requirements and integration into clinical workflows and demonstrate the potential clinical impact for therapies like transcranial magnetic stimulation (TMS). By leveraging Mask R-CNN in the neuroimaging domain, this work represents a novel step toward personalized neuromodulation and improved treatment outcomes in psychiatry.