Fast and Robust Unsupervised Identification of MS Lesion Change Using the Statistical Detection of Changes Algorithm

Abstract

We developed a robust automated algorithm called statistical detection of changes for detecting morphologic changes of multiple sclerosis lesions between 2 T2-weighted FLAIR brain images. Results from 30 patients showed that statistical detection of changes achieved significantly higher sensitivity and specificity (0.964, 95% CI, 0.823-0.994; 0.691, 95% CI, 0.612-0.761) than with the lesion-prediction algorithm (0.614, 95% CI, 0.410-0.784; 0.281, 95% CI, 0.228-0.314), while resulting in a 49% reduction in human review time (P = .007).