Currently, a wide array of magnetic resonance imaging (MRI) protocols are pre-defined for different diseases. For example, in neurological diseases, there are protocols for epilepsy, stroke, multiple sclerosis, etc. Under each disease category, the protocols are further divided into pediatric or adult patients, and each protocol consists of a collection of MRI sequences aiming to obtain different tissue contrasts, orientations, imaging resolution, and coverage. Furthermore, some of the protocols include the injection of MRI contrast agents and acquisition of additional images post-contrast.
When a diagnostic MRI is ordered the protocol may be selected by a technologist or ordering provider, with a relatively small menu of protocols (e.g. routine brain, pituitary, tumor, stroke protocols). The ordering provider usually has limited knowledge about the scanner protocols, while the technologist usually has limited knowledge of the medical history and clinical indication of the patient. This approach lacks consistency across the clinical spectrum, from patients to the radiologists, and is not specific for each patient, even in the same disease category. Additionally, it does not allow the opportunity to alter the patient’s diagnosis during the imaging session, which would be beneficial for a patient undergoing a brain MRI based on a headache and confusion that reveals a brain mass, suggesting a possible brain tumor. This approach can also result in the need to re-image a patient, which is costly and wastes valuable resources for the patient and clinical center.
Researchers in the Department of Radiology and the Biomedical Research Imaging Center at the University of North Carolina at Chapel Hill have developed a smart MRI protocol to optimize imaging for each patient. The smart protocol utilizes machine learning to eliminate the need for a radiologist or technologist to select a protocol for each patient and instead to automatically selects an initial protocol for each patient based on their clinical history or tissue/organ of interest. Following acquisition of the initial images and analysis by the machine learning platform, coupled with the patient’s clinical history, the protocol is adjusted and additional image sequences are used for the second set of images. These steps are repeated until the desired images are collected for the patient. This ensures consistency across patients with the same clinical indications, determines if a contrast agent is appropriate, and if the scan should be stopped early due to movement.
Individualized protocol for each patient to maximize diagnostic accuracy, and minimize risk and time spent in scanner.
Minimizes need for re-imaging, leading to cost and resource savings.
Automated protocol/sequence selection will lead to more efficient and uniform scanning of patients with similar clinical indications.
Optimal risk/benefit for patients by determining the most efficient use of contrast agents.
In addition to smart protocols for MRI, the approach could be more widely applied for development of smart protocols for other imaging modalities, such as PET, CT, and ultrasound, as these are often combined when diagnosing a patient.