Oriented Object Detection Deep Learning Model for Automated Prescription of 3D MR Spectroscopic Imaging of the Brain

Motivation: The expertise required to acquire high-quality Magnetic Resonance Spectroscopic Imaging (MRSI) data is a barrier to its adoption into routine clinical practice. Goal(s): To develop a machine learning-based automated technique for the prescription of 3D MRSI examinations. Approach: We developed an ML-based pipeline to generate an oblique excitation volume based on MR images of the brain. The models were trained on a dataset of 714 MRI/MRSI exams of patients with brain tumors and evaluated on a validation set. Results: The models achieved a mean intersection over union (IOU) of 0.85 (axial) and 0.75 (sagittal) with standard deviations of 0.07 and 0.09 respectively. Impact: This study demonstrates the feasibility of a machine learning-based automated prescription of 3D MRSI acquisition of the brain. This would help solve a major challenge of incorporating metabolic imaging into the routine clinical care of patients with brain tumors.