MIDI (MR Imaging Abnormality Deep Learning Identification)

An automated strategy for identifying abnormalities in head scans could address the unmet clinical need for faster abnormality identification times, potentially allowing for early intervention to improve short- and long-term clinical outcomes. Radiologist shortages and increased demand for MRI scans lead to delays in reporting, particularly in the outpatient setting.

Furthermore, there is a wide variation in the management of incidental findings (IFs) discovered in 'healthy volunteers.' The routine reporting of 'healthy volunteer' scans by a radiologist poses logistical and financial challenges. It would be valuable to devise automated strategies to reliably and accurately identify IFs, potentially reducing the number of scans requiring routine radiological review by up to 90%, thus increasing the feasibility of implementing a routine reporting strategy.

Deep learning is a novel technique in computer science that automatically learns hierarchies of relevant features directly from the raw inputs (such as MRI or CT) using multi-layered neural networks. A deep learning algorithm will be trained on a large database of head MRI scans to recognize scans with abnormalities. This algorithm will be trained to classify a subset of these scans as normal or abnormal and then tested on an independent subset to determine its validity.

If the tested neural network demonstrates high diagnostic accuracy, future research participants and patients may benefit, as not all institutions currently review their research scans for incidental findings and clinical scans may not be reported for weeks in some cases. In both research and clinical scenarios, an algorithm could rapidly identify abnormal pathology and prioritize scans for reporting.

In summary, the aim is to develop a deep learning abnormality detection algorithm for use in both research and clinical settings.