Experimental Evaluation of Brain Tumor Image Segmentation and Detection Using CNN Model

Abstract

Listen

Translate article

AbstractIn this study, the brain tumor detection technique using the digital image is discussed. There are two ways to perform the detection: one is segmentation and the other is classification. The segmentation process used the adaptive threshold by which the areas of interest are first extracted from the digital image, and also reliable segmentation is engineered by a model-based approach, i.e., modified Markov random field (MRF). The classification process used the MRF segmented areas that are sorted into regular and suspicious and used a method called Convolution Neural network (CNN). In this chapter, the dataset is divided into two parts and it consists of a set of screen/films of the brain with abnormal (105) brain images and normal (167) brain images that have been tested. Proven that there are biopsies of 48, with the malignant mass of different types, and subtlety are contained later. A free-response receiver characteristic operating curve is used to detect the algorithm accuracy which demonstrates the association between the true positive massed detection and the variety of false-positive alarms per image. The outcome indicated that a sensitivity of 90% can be carried out in the prediction of various types of masses at the expense of false detection of two signals per image. The algorithm becomes notably successful in the prediction of the nominal cancers demonstrated by SPL les/masses/size of 10 mm. In the dataset of 16 cases, a sensitivity of 94% was observed with 1.5 false alarms per image. An extensive analysis of the consequences of the algorithm’s parameters on its sensitivity and specificity became additionally carried out to optimize the technique for a medical, observer overall performance analysis.KeywordsA brain tumorSegmentationConvolutional neural network (CNN)EpochsHealthcareImage augmentation