An Intelligent method for MS diagnosis from brain MRI images using image segmentation and bee colony optimized LSTM network

Abstract

Traditional methods of estimating lesion volume from MRI images are based on visual inspection, mainly human intervention using available window-level presets. The aforementioned methods for visually examining images generally imply referring to the practice of scanning X-ray images with the naked eye or some optical magnifier. Such practices highly consume time and have notable inaccuracies implicit. Therefore, the design of image processing and machine learning-based error-correction systems in diagnosing diagnostic errors due to human misinterpretation is very much essential. One of such recently more prevalent activities, considered by many to be a disease, is multiple sclerosis (MS). If this disease is not diagnosed at an early stage, it will lead to death at a young age with multiple complications. This paper proposes an accurate intelligent system for diagnosing MS. If morphological operations are optionally acted upon the lesion areas after histogram equalization has enhanced the quality of MRI images in the pre-processing phase, a large role of segmenting lesion areas may easily be undertaken by morphological operations in the next stages. The diagnosis thus flows to be about regions on the sample, thereby trying adoptions of samples with similar features into one class. The preferable count of hidden neurons in the LSTM network is fine-tuned by the Bee Colony Optimization algorithm to enhance the classification accuracy. The proposed technique finally attains an accuracy of 99.6%.