MIT researchers trained a machine learning system on MRI data from patient’s suffering from neurodegenerative diseases. By leveraging this vast pool of data, the team found that they were able to cut the prediction error rate with new patients from 20% to 10%, a reduction of 50%. The data that they used to train the machine was taken from a longitudinal study of MRI’s taken on the same subjects months and years apart.
Aside from being able to gauge the possibility of neurological disease on individuals up to 10 years out, the technology also holds great importance for experimental Alzheimer’s drug development. With this methodology, drug-makers can more accurately gather data and make predictions on MRI data alone, which is much cheaper than current processes that require MRI and PET scans. “People think MRI is expensive, but it’s only a fraction of what PET scans cost, If machine-learning tools can help avoid the need for PET scans in evaluating patients early in the disease course, that will be very impactful,” says Bruce Rosen of Harvard.