[O-3-305] CAD system of intracranial aneurysms at MRA images with intensity inhomogeneity correction
A computer aided diagnosis (CAD) system of asymptomatic unruptured intracranial aneurysms has been developed based on a three-dimensional (3D) blob structure enhancement (BSE) filter on 3.0 T magnetic resonance angiography (MRA). However, the number of false positives was 5.8 with a sensitivity of 100% for 10 cases in a leave-one-out-by-patient test method. In this study, an intensity inhomogeneity correction (IIC) step, which may affect the calculation of image features, has been incorporated into the CAD system. The IIC employed the unified segmentation framework, whose model was based on a mixture of Gaussians, and was extended to incorporate a smooth intensity variation and nonlinear registration with tissue probability maps. The 10 MRA images (matrix size: 512 x 512 x 224-258, voxel size: 0.35 mm) with 10 unruptured aneurysms (diameter: 2.4 to 8.4 mm, mean: 6.6 mm) were used in this study. The initial aneurysm candidates were identified by use of a multiple gray-level thresholding technique and a region growing technique with monitoring certain image features on the BSE images. A number of false positives were reduced by using different rules on 17 image features associated with morphology and gray levels. A support vector machine with a third order polynomial kernel, which can learn the image features of true positives and false positives, was used for further removal of false positives. The CAD system detected all aneurysms while reducing false positives per patient from 5.8 to 3.6 by incorporating the IIC step into the CAD system. This CAD system could be feasible for assisting radiologists in detection of asymptomatic unruptured intracranial aneurysms in 3.0 T MRA.