Thursday, December 4 | 9:30 a.m.-9:40 a.m. | SSNR13-1 | Room N228
In this award-winning presentation, researchers will show how a deep-learning model can be useful in diagnosing multiple sclerosis (MS).
In research that received the Kuo York Chynn Neuroradiology Research Award, doctoral student Jiajian Ma from New York University will present a deep-learning model for predicting MS using brain structural MRI (sMRI) and diffusion MRI (dMRI).
The model was developed to extract visual features from multimodal MRI scans, produce modality-specific saliency maps predicting MS at the voxel level, and then fuse them into a single subject-level MS probability, according to Ma and colleagues.
The project involved an internal dataset of 7,651 brain MRI studies from 6,564 patients scanned between 2014 and 2020 at a multicenter health system in New York. Ma and colleagues included 1,333 MS cases (per 2010/2017 McDonald criteria) and 2,484 non-MS cases.
The team split data at the patient level into training (n = 6,432), validation (n = 331), and test (n = 888; 467 with both 3D FLAIR and dMRI). For validation, they used an external test set of 490 patients (70 MS, 420 non-MS with white-matter lesions) from four public datasets. Model performance was evaluated using receiver operating characteristic area under the curve (ROC-AUC) and partial AUC (pAUC) with a maximum false-positive rate of 5%.
On the internal test set of 467, including 95 MS patients, the model using 3D FLAIR alone achieved an ROC-AUC of 0.957, according to the group. These and other findings provide preliminary evidence for incorporating diffusion MRI into routine MS imaging protocols, they noted.
"Deep learning models can accurately differentiate MS from other white matter lesions using structural MRI, with strong generalization to external datasets," they wrote. "Incorporating diffusion MRI further improves prediction performance, especially at clinically relevant specificity levels (false-positive rate <5%)."
What's more, the study results offer preliminary evidence for incorporating diffusion MRI into routine MS imaging protocols, according to the authors.
Get all of the details at this Thursday morning presentation.
