The development of CAD system for hemorrhagic stroke in computed tomography images

Abstract

The purpose of this study is to develop a hemorrhagic stroke computer-aided detection system to quantify three values in computed tomography, including values of brain midline shift (BMS), bleeding volume (BV) and edema volume (EV), which are basic index for physicians to observe. The median filter and region growing were used to remove the noise and skull stripping in single slice and then the brain tissue's shift axis was detect based on the location of the ventricle and pineal gland. The bleeding region and edema region were segmented by checking the symmetrical of left and right brain. Finally, after calculate the volume through each slice these three index (BMS, BV, and EV) can be obtained. Phantoms and clinical CT images which including 240 slices images from 8 cases, were used to verify this system. The error of midline shift was less than 5% and the volume difference is less than 3.47% for phantom images in average. After comparing the area of bleeding region (ABR) and edema region (AER) with this system calculated and doctor selected in those CT, the results showed that the average difference of ABR is 8.8±3.13%, the average of degree of coincidence is 86.4±5.43%, and the average difference of AER is 14.1±4.48%, the average of degree of coincidence is 77.4±6.58%. Thus, the proposed method can clearly benefit not only visual diagnostics, but also quantitative methodologies about hemorrhagic stroke-related parameters.