Unsupervised Segmentation of MR Images for Brain Dock Examinations

摘要

As described herein, we propose an unsupervised segmentation method for magnetic resonance (MR) brain imaging by hybridizing the self-mapping characteristics of 1-D Self-Organizing Maps (SOMs) and by using incremental learning functions of fuzzy Adaptive Resonance Theory (ART). The proposed method requires no operator to specify the representative points. Nevertheless, it can segment the tissues (e.g., cerebrospinal fluid, gray matter, and white matter) necessary for brain atrophy diagnosis. To evaluate the effectiveness of the proposed method, we specifically examine Fuzzy C-means (FCM) and Expectation Maximization Gaussian Mixture (EM-GM) with prior setting of the cluster number, and Mean Shift (MS) without prior setting of the cluster number. These experiments on the two metrics confirmed that our method can achieve higher accuracy than these conventional methods. Additionally, we propose a Computer-Aided Diagnosis (CAD) system for use with brain dock examinations based on case analysis of diagnostic reading. We construct a prototype system to reduce loads to diagnosticians during quantitative analysis of the degree of brain atrophy. Through field testing of 193 examples from brain dock examinations, we also demonstrate the possibility of efficiently supporting diagnostic work in the clinical field because the alternation of brain atrophy attributable to aging can be quantified easily irrespective of diagnosticians’ subjectivity.