A Deep Learning-based Assessment Tool for the Objective Evaluation of Intracranial MRI Quality in Stereotactic Radiosurgery

Motivation: GK-SRS is a treatment modality capable of delivering radiation with sub-millimeter precision, requiring high-quality MRI for accurate target delineation. Motion artifacts often compromise image quality, affecting treatment accuracy. Goal(s): To develop an assessment tool for the objective evaluation of intracranial MRI quality in GK-SRS, ensuring consistent image quality for treatment planning. Approach: A 3D CNN was trained on a set of clinical images, supplemented with images featuring artificially generated motion artifacts. Images were categorized by a team of medical professionals on quality. Results: The model achieved 82.5% accuracy on a validation set of 359 clinical images. All clinically unacceptable images were successfully identified. Impact: Due to the precise nature of Gamma Knife Stereotactic Radiosurgery, accurate target delineation is crucial. The development of an objective intracranial MRI evaluation tool will contribute to ensuring consistent image quality in treatment planning, thereby improving clinical outcomes.