Support vector machines based analysis of brain SPECT images for determining cerebral abnormalities in asymptomatic diabetic patients

Purpose: An image processing method was developed to investigate whether brain SPECT images of patients with diabetes mellitus type II (DMII) and no brain damage differ from those of normal subjects. Materials and methods: Twenty-five DMII patients and eight healthy volunteers underwent brain 99mTc-Bicisate SPECT examination. A semi-automatic method, allowing for physician's interaction, was developed to delineate specific brain regions (ROIs) on the SPECT images. Twenty-eight features from the grey-level histogram and the spatial-dependence matrix were computed from numerous small image-samples collected from each specific ROI. Classification into 'diabetics' and 'non-diabetics' was performed for each ROI separately. The classical least squares-minimum distance (LSMD) classifier and the recently developed support vector machines (SVM) classifier were used. System performance was evaluated by means of the leave-one-out method; one sample was left out, the classifier was trained by the rest of the samples, and the left-out sample was classified. By repeating for all samples, the classifier's performance could be tested on data not incorporated in its design. Results: Highest classification accuracies (LSMD: 97.8%, SVM: 99.1%) were achieved at the right occipital lobule employing two features, the standard deviation and entropy. For the rest of the ROIs classification accuracies ranged between 84.5 and 98.6%. Conclusion: Our findings indicate cerebral blood flow disruption in patients with DMII. The proposed system may assist physicians in evaluating cerebral blood flow in patients with DMII undergoing brain SPECT.