Computer Aided Diagnosis System of Brain Tumors in PET/CT Images

Abstract

In this paper, we proposed a computer aided diagnosis system (CADS) that supports to detect the existence of tumor in the area of the brain from a given FDG-PET/CT image. Firstly, tilt correction was applied to brain image based on a computation of the mid-sagittal plane in the head. Secondly, tumor detection was performed by detecting asymmetrical regions between images of the left and right cerebral hemispheres. In this step, an improved SURF algorithm was utilized to detect the asymmetrical regions between the left and right sides in a 2D axial slice of the brain image. In cases where the 2D axial slice data was identified as symmetrical, one supplementary method based on typical illegal shape identification was applied so that the symmetrical tumor could be detected. 80 cases (10 cases with brain tumors and 70 normal cases) were tested in this experiment. The results showed this system could detect brain tumors effectively and was capable of identifying a health brain pattern and mostly avoided the incorrect identification of tumors.