Automatic detection of white matter hyperintensities via mask region-based convolutional neural networks using magnetic resonance images

Abstract

Because white matter hyperintensities (WMHs) are associated with many different types of brain disease or disorders, they need to be detected as early as possible. Accurate detection of WMHs occurring in the brain is important for physicians to decide on the appropriate treatment method and to determine the type, location, size, and boundary detection of the pathologic case with high accuracy. This study proposes a mask region-based convolutional neural network method for the automatic detection of WMHs on magnetic resonance (MR) scans. Three datasets, one of which is specific to this study and two of which are given publicly available, are provided for experimental studies. As a result of test set in the study, multiple sclerosis lesions and brain tumors are successfully detected on MR slices with a high mean average precision score of 0.94. In addition, precision and the Dice similarity coefficient have scores as 0.86 and 0.82, respectively. © 2022 Elsevier Inc. All rights reserved.