Deep Learning-Based Prediction of PET Amyloid Status Using MRI.

background and purposeIdentifying amyloid-beta (Aβ)-positive patients is essential for Alzheimer disease clinical trials and disease-modifying treatments but currently requires PET or CSF sampling. Previous MRI-based deep learning models using only T1-weighted (T1w) images have shown moderate performance. MATERIALS AND

methodsMulticontrast MRI- and PET-based quantitative Aβ deposition were retrospectively obtained from 3 public data sets: ADNI, OASIS3, and A4. Aβ positivity was defined using the recommended Centiloid threshold of each data set. Two EfficientNet models were trained to predict amyloid-positivity: one by using only T1w images and another incorporating both T1w and T2 FLAIR. Model performance was assessed using an internal held-out test set, evaluating area under the curve (AUC), accuracy, sensitivity, and specificity. External validation was conducted using an independent cohort from Stanford Alzheimer Disease Research Center. DeLong and McNemar tests were used to compare AUC and accuracy, respectively.

resultsA total of 4056 examinations (mean age: 71.6 [SD, 6.3] years; 55% female; 55% amyloid-positive) were used for network development, and 149 examinations were used for external testing (mean age: 72.1 [SD] 9.6] years; 57% female; 56% amyloid-positive). The multicontrast model outperformed the single-technique model in the internal held-out test set (AUC: 0.67; 95% CI, 0.65-0.70;

conclusionsThe use of multicontrast MRI, specifically incorporating T2 FLAIR in addition to T1w images, significantly improved the predictive accuracy of PET-determined amyloid status from MRIs by using a deep learning approach.