Computer-Aided Diagnosis Scheme for Detection of Unruptured Intracranial Aneurysms in MR Angiography

The detection of unruptured intracranial aneurysms is a major subject in magnetic resonance angiography (MRA) images. However, it is difficult for radiologists to detect small aneurysms on the maximum intensity projection (MIP) images, because adjacent vessels overlap with the aneurysm. The purpose of this study was to develop an automated computerized detection of aneurysms in order to assist radiologists' interpretation as a "second opinion." The vessels were first segmented from background by use of gray-level thresholding and region growing technique. The gradient concentrate (GC) filter was then applied to the segmented vessels for enhancement of aneurysm. The initial aneurysm candidate was identified in the GC image with a gray level threshold. For removal of false positives (FPs), we determined three features, i.e., size, sphericity, and mean value of GC image in each of the candidate regions. Finally, the rule-based schemes with these features and quadratic discriminant analysis were applied for distinction between aneurysms and FPs. The sensitivity of our method for detection of aneurysms was 100% (7/7) with 1.85 FPs per patient. With our computerized scheme, all aneurysms were detected correctly with low FP rates, and would be useful in assisting radiologists for identifying correct aneurysms and for reducing the interpretation time