Study: MRA + computer-aided detection can better detect brain aneurysms

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
Image source: Chestnut Medical Technologies

MR angiography (MRA), with the addition of a specially designed computer-aided detection (CAD) scheme, has proven to have greater accuracy than MRA alone, and may have the ability to improve the sensitivity of intracranial aneurysm detection on MR images, according to the findings of a collaborative project of Mayo Clinic and IBM.

The results of the project--which were published in the February edition of the Journal of Digital Imaging--showed that while MRA has emerged as a useful non-invasive method for aneurysm detection, most neuroradiologists have poor sensitively to small (less than 5mm) aneurysms on MR images, with a detection rate between 30 to 60 percent on average.

A patient suspected of having a brain aneurysm would typically undergo invasive tests using a catheter and injectable dye, which carries risks of neurological complications. Noninvasive MRA imaging, while becoming more widely-used for this patient population, noted the authors, still needs improvement. However, with the implementation of a specific CAD system, the researchers found that the rate of detection rose to 95 percent overall.

The CAD scheme, which resulted from a collaboration of the Mayo Clinic and IBM, uses technology that combines two complementary techniques. First, the scheme locates points of interest on individual MRA datasets, by segmenting the intracranial arteries. The scheme also identifies points of interest directly from the unsegmented image dataset. A small fraction of points of interest are then taken and labeled as aneurysm candidates from the collected points of interest after a series of calculations and predetermined rules are applied by the scheme.

Bradley Erickson, MD, the study's senior author and co-director of the Medical Imaging Informatics Innovation Center at the Mayo Clinic, noted that the technology highlights likely aneurysms on MR images to enhance identification and detection by the radiologist before the aneurysms can result in neurological damage or brain hemorrhage.

The project was put into practice on 287 datasets containing 147 aneurysms, which were verified with digital subtraction angiography. Following the scan, the images were shipped to servers at Mayo and IBM at the Medical Imaging Informatics Innovation Center located on the Mayo campus in Rochester, Minn. Once the CAD scheme was applied and potential aneurysms were marked--including those less than 5mm--the images were sent to a group of radiologists for further analysis.

The authors wrote that for two different operating points, the CAD system achieved a sensitivity rating of 80 percent and 95 percent, respectively, with three false positives per case on average (71 percent for aneurysms less than 5mm, with nine false positives per case, on average).

“According to the experimental results,” wrote Erickson and colleagues, “the sensitivity of our CAD scheme for aneurysm [detection] was excellent. Indeed, even using the CAD without human input, sensitivities were substantially higher than previous reports (e.g., a sensitivity range of 69 to 99 percent). “Furthermore, and potentially more significant, is that for small aneurysms, the sensitivity of our CAD scheme can be up to 91 percent, which is much higher than the reported sensitivity of 35 to 56 percent by human detection.”