Unsupervised classification of brain tissue density distribution in CT scans

Medical application of a computerized system suitable for the detection of minimal tissue density variation can be useful in the analysis of brain scans in subjects affected by different pathologies. Quantification of the severity of the disease through estimation of the lesions is vital for the diagnosing, understanding and monitoring of the illness and its treatment. In this paper, we present an unsupervised approach of a fully automated segmentation and classification process for digital medical images. The proposed scheme is based on an integrated approach, which employs a statistical method for feature extraction and a Self-Organizing Map for classification. The developed architecture was trained, tested, and evaluated on both normal and pathological cases taken from CT brain scans. During the development of our method, we implemented a fast and easy technique for computing the statistical features. The segmentation algorithm produces reliable results thereby making our scheme suitable for practical applications and useful in the design of a totally unsupervised approach for the diagnosis of CT scans.