An OTP Framework for effective tumor segmentation and analysis in brain MRI images

Abstract

Analysis and segmentation of tumor structure in brain MRI images is considered as trial and complicated task due to the unpredictability characteristics of tumor region and the difficulty in computing the accurate tumor size, tumor texture and tumor location. In this work, an Optimal Three-Phase (OTP) MRI segmentation framework is proposed for an effective interpretation and classifying the varied MRI dataset for further investigation and improved results. The OTP framework brings diverse popular segmentation frameworks to undertake MRI investigation process. An Optimal Three-Phase (OTP) MRI segmentation framework is intended to assortment the services of most popular FCM clustering framework, Watershed Segmentation framework and EMG (Expectation Maximization and Gaussian) mixture framework for an automatic and precise analysis and diagnosis of tumor in MRI. The proposed Optimal Three-Phase (OTP) Framework has a comprehensive tumor analysis structure which comprise of three phases for identification, segmentation and extraction of disorder region in MRI. The OTP framework has automatic identification of disorder in MRI images consecutively tumor region segmentation. The first level deals by the manifestation and identification of brain tumor in input MRI datasets. The second level deals with segmentation of tumor and analysis of tumor region in MRI images. The third level is constructed on tumor region extrapolation which is capable to extract the tumor core, active tumor and whole tumor region. The standard input MRI dataset comprises of 1000 patients brain MRI images which is composed of T1C MRI images form BraTS. The assessment results of proposed framework for analysis tumor achieved an accuracy of 98.23% ± 1.1% compared with existing techniques. These favorable outcomes demonstration that the offered framework is most efficient and robust with respect to conventional classification techniques in terms of efficient results on standard BraTS dataset.