| Abstract |
: |
In this paper, a brain lesion detection and classification approach using thresholding and a rule-based classifier is proposed. Four types of brain lesions based on diffusion-weighted imaging i.e. acute stroke, solid tumor, chronic stroke, and necrosis are analyzed. The analysis is divided into four stages: pre-processing, segmentation, feature extraction, and classification. In the detection and segmentation stage, the image is divided into 8x8 macro-block regions. Adaptive thresholding technique is applied to segment the lesion’s region. Statistical features are measured on the region of interest. A rule-based classifier is used to classify four types of lesions. Jaccard’s similarity index of the segmentation results for acute stroke, solid tumor, chronic stroke, and necrosis are 0.8, 0.55, 0.27, and 0.42, respectively. The classification accuracy is 93% for acute stroke, 73% for solid tumor, 84% for chronic stroke, and 60% for necrosis. Overall, adaptive thresholding provides high segmentation performance for hyper-intensity lesions. The best segmentation and classification performance is achieved for acute stroke. The establishment of the technique could be used to automate the diagnosis and to clearly understand major brain lesions. |
