An Automated Framework for MRI Protocol Management and Clinical Exam Monitoring
Abstract
Purpose
To establish an automated framework for MRI protocol management and clinical examination monitoring that enables scalable and queryable protocol representation, automated exam-to-protocol classification, real-time examination performance analysis.
Methods
The proposed system consists of three primary components. (1) Protocol conversion: Metadata from a reference MRI examination are transformed into a structured JSON-based protocol representation. Each protocol file contains protocol-level attributes (scanner, protocol, and creation date) supporting indexing and querying and a collection of sequence objects, each encapsulating standard and special acquisition parameters and derived variables such as voxel size and relative signal-to-noise ratio (SNR). (2) Exam classifier: Each examination is represented as a set of sequences characterized by eight vendor-neutral parameters. Summary statistics (mean, max, min, and standard deviation) are computed across the sequence set and concatenated into a fixed-length feature vector representing the entire exam. A multinomial logistic regression model is trained on standardized feature vectors to classify exams into predefined protocols using cross-entropy loss with L2 regularization. (3) Sequence labeler: The training exams are first decomposed into individual sequences. Each sequence is represented by a fixed-length feature vector composed of a subset of acquisition parameters. A regularized multinomial logistic regression model is trained to map feature vectors to predefined sequence labels.
Results
A total of 228 training examinations were collected across three brain protocols (metastasis, surgical-planning, and tumor) from three different scanners. The dataset was separated into the training set (120 exams) and the testing set (108 exams). All test exams were correctly classified into their respective protocols, and all sequences were correctly labeled. Relative SNR performance was assessed within the scanner–protocol groups, and outliers were identified and compared against the standard protocols.
Conclusion
This automated, metadata-driven framework enables robust MRI protocol management, examination monitoring, and performance assessment, demonstrating potential for operational oversight in clinical MRI workflows.