EFFICIENT SEGMENTATION MODEL USING MRI IMAGES AND DEEP LEARNING TECHNIQUES FOR MULTIPLE SCLEROSIS CLASSIFICATION

Abstract

The segmentation models employing deep learning offer successful outcomes over multiple medical image complex data resources and public data resources important for huge pathologies. During the identification of multiple sclerosis, the observation of entire tumors from the magnetic resonance imaging (MRI) sequence is complex. Furthermore, it is necessary to identify the small tumors from the images in the prognosis phase to offer good treatment. The deep learning-assisted identification models solve the issue of the imbalance data, and the false positive results are more in the conventional models. Besides, these methodologies offer a good tradeoff between the precision measure and recall measure. Thus, the latest deep learning-assisted MRI image segmentation and categorization model is developed to detect multiple sclerosis at the initial stage. In this study, the MRI images are initially gathered. The gathered images are directly given to the image segmentation process, where the Multi-Scale Adaptive TransResunet++ (MSAT) is adopted to perform the lesion segmentation appropriately. The attributes present in the MSAT are optimized with the support of the developed random opposition of cicada swarm optimization (ROCSO). Then, the segmented pictures are subjected to the categorization process where the hybrid and dilated convolution-based adaptive residual attention network (HDCARAN) is utilized. The HDCARAN categorizes the lesions from the MRI images very effectively and detects the multiple sclerosis of patients. Here, the attributes present within the HDCARAN are tuned via the same ROCSO. The implementation results are analyzed through the previously developed multiple sclerosis detection schemes to evaluate the effectiveness of the designed model with respect to several functionality measures. The implementation of HDCARAN results in the detection of lesions in the initial stages, with a 94% specificity level, as well as efficiency performance based on existing models. The model also obtains a 94.5% accuracy, as opposed to the other conventional systems. The results show the proposed method performs better than other machine learning models.