Deep-learning model outperformed neuroradiologists in predicting risk for disease development.
By adding an artificial intelligence (AI) tool into the mix, researchers were able to use brain MRI scans to predict a patient’s risk level for developing Alzheimer’s disease, according to newly published research.
The risk assessment also included other, non-imaging factors, such as gender, age, and cognitive test scores. Results were published in the journal Brain.
A Boston University team, led by Shangran Qui, Prajakta Joshi, Matthew Miller, and Chonghua Zue, used the Alzheimer’s Disease Neuroimaging Initiative (ADNI) to teach a deep-learning algorithm. Based on performance, not only did they determine the model was substantially accurate on three distinct datasets, but they also found that it performed better than neuroradiologists.
“This framework provides a clinically adaptable strategy for using routinely available imaging techniques, such as MRI to generate nuanced neuroimaging signatures for Alzheimer’s disease diagnosis, as well as a generalizable approach for linking deep learning to pathophysiological processes in human disease,” the team wrote.
The team used scans from 414 patients who had been diagnosed with Alzheimer’s, as well as control patients from the ADNI. By analyzing T1-weighted brain MRI images, the fully convolutional neural network can generates color-coded disease probability maps, and a multi-layer perceptron evaluates those maps in order to create a binary Alzheimer’s disease status classification.
Investigators assessed their model on the ADNI dataset. They also did so on three independent patient groups – 382 patients from the Australian Imaging, Biomarker, and Lifestyle Flagship Study of Aging, 102 patients from the Framingham Heart Study, and 582 individuals from the National Alzheimer’s Coordinating Center.
According to their results analysis, the team determined the most effective and accurate model also included evaluations based on a patient’s gender, age, and score on the Mini-Mental State Examination. Compared to the evaluation results of 11 neuroradiologists who read 80 randomly chosen ADNI scans, the investigator’s model performed better, producing an area under the curve of 0.996 compared to the neuroradiologists’ 0.920. Additionally, the model was also highly accurate in identifying high-risk cerebral regions.
Additional testing is needed to validate whether this model can be used to improve clinical care and outcomes, the team wrote, but it has the potential to expand neuroimaging tools and techniques available for disease detection and prevention.
Stay at the forefront of radiology with the Diagnostic Imaging newsletter, delivering the latest news, clinical insights, and imaging advancements for today’s radiologists.
Large Medicare Study Shows Black Men Less Likely to Receive PET and MRI for Prostate Cancer Imaging
August 3rd 2025An analysis of over 749,000 Medicare beneficiaries diagnosed with prostate cancer over a five-year period found that Black men were 13 percent less likely to receive PET imaging and 16 percent less likely to receive MRI in comparison to White men.
The Reading Room Podcast: Current and Emerging Insights on Abbreviated Breast MRI, Part 3
August 3rd 2025In the last of a three-part podcast episode, Stamatia Destounis, MD, Emily Conant, MD and Habib Rahbar, MD, share additional insights on practical considerations and potential challenges in integrating abbreviated breast MRI into clinical practice, and offer their thoughts on future research directions.
Reducing the Interval Breast Cancer Rate of Screening DBT: Can AI Have an Impact?
August 3rd 2025In a retrospective review of screening digital breast tomosynthesis (DBT) exams for over 200 women with interval breast cancers, researchers found that AI provided accurate localization of cancers in 32.6 percent of the cases.
The Reading Room Podcast: A Closer Look at Remote MRI Safety, Part 3
August 3rd 2025In the third of a three-part podcast episode, Emanuel Kanal, M.D. and Tobias Gilk, MRSO, MRSE, discuss strategies for maintaining the integrity of time-out procedures and communication with remote MRI scanning.