Tomohiko Masumoto MD, Presenter: Nothing to Disclose

Yoshitaka Masutani PhD, Abstract Co-Author: Nothing to Disclose

Naoto Hayashi MD, Abstract Co-Author: Nothing to Disclose

Shigeki Aoki MD, Abstract Co-Author: Nothing to Disclose

Harushi Mori MD, Abstract Co-Author: Nothing to Disclose

Osamu Abe MD, PhD, Abstract Co-Author: Nothing to Disclose

Kuni Ohtomo MD, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

Magnetic resonance angiography (MRA) is widely used for screening cerebral aneurysms. However, detection of aneurysm in a small size is difficult and time-consuming for radiologists. Based on volumetric image analysis, we developed an automated detection system for cerebral aneurysm in MRA, and evaluated this system with clinical data obtained by MR scanners provided by several vendors.

MRA volume data was acquired by conventional time-of flight method.The software for automated detection was developed on a PC workstation. Segmentation of vascular structures was performed by thresholding, connected　component analysis and morphological analysis. In segmented voxels, features including gradient, second derivatives, and curvature were calculated to determine the aneurysm-candidate voxels. The candidates were sorted by the volume, and overlayed on a 3D volume-rendering image.As the first experience, we applied this automated detection system to 20 clinical cases (12 positive, 8 normal) acquired by the fixed protocol in a single scanner. We evaluated feasibility and detection performance. To evaluate flexibility of the system, we also examined the MRA data sets acquired by the other protocols with various MR scanners.

In all cases, the detection process was completed automatically. The entire processing time was short enough (28 seconds in average) for clinical use by calculation of features in only segmented vascular structures.All the true positive detections were identified in higher ranks than false positive detections. With FROC analysis, a sensitivity of 100% was achieved with an average of 1.85 false positive detections per case.Because MRA data acquired by the other protocols had different properties of signal intensity, adjustment of parameters was needed during segmentation and feature analysis. With appropriate adjustment, our system was able to handle MRA data sets of various protocols and scanners.

We developed the automated detection system which enabled fast detection of cerebral aneurysm in MRA with excellent sensitivity. This system is expected to improve efficiency of radiologists in the diagnosis of cerebral aneurysm.

Masumoto, T, Masutani, Y, Hayashi, N, Aoki, S, Mori, H, Abe, O, Ohtomo, K, et al, , Automated Detection System for Cerebral Aneurysm in Magnetic Resonance Angiography: Development and Initial Clinical Experiences in Multiple MRI Scanners.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4410511.html