IoHT with Cloud-Based Brain Tumor Detection Using Particle Swarm Optimization with Support Vector Machine

Introduction

Rapid development in data and micro-electromechanical system (MEMS) methods results in the deployment of the Internet of Things (IoT), which enables objects, data, and virtual environments to interact with each other [I]. Various fields apply IoT in data collection tasks such as transport, homes, hospitals, cities, etc. The exponential development of IoT-dependent healthcare tools and sensors exist in real time [2]. With an increased cost of medications as well as the existence of diverse diseases all over the world, it is significant for healthcare from a hospital-based structure to a patient-centric structure. To control the disease, the ubiquitous sensing capabilities derived from IoT devices were applied to detect the possibilities of developing the disease for a user. The interconnection of IoT and CC is assumed to be more applicable in monitoring affected people in remote areas by providing enough support for physicians [3]. IoT has been provisioned by the application of virtual unconstrained utilities as well as resources of cloud computing (CC) to maintain technical shortcomings such as storage, processing, and power. Simultaneously, CC provides the merits of IoT under the expansion of its value to deal with real-time applications and to provide massive facilities in a distributed as well as dynamic fashion. Therefore, IoT and CC could be applied for developing novel applications and services in the healthcare domain [4].

IoHT is an alternate combination of IoT and healthcare, which has been deployed in the healthcare sector [5]. In the case of massive IoT domains, the major duty of big data analytics as well as CC is a popular methodology. Ma et al. [6] proposed a backend structure that activates cognitive facilities in healthcare recommending that a cloud approach must not be basically homogeneous and offers medical data transfer and CC service layers. The ВТ affects people at all age groups and it results in increased death rate [7]. A tumor is comprised with tissues of anomalous or abnormal cells. Also, it is associated with a benign ВТ. which is noncancerous and does not spread to neighboring tissues; however, it is malicious and can cause death.

An alternate case named malignant BTs is referred to as cancerous, which is developed in the brain and uniquely breeds when compared with benign tumors and spreads in nearby tissues. Magnetic resonance imaging (MRI) has been applied to gain knowledge of tumors and calculate the spreading value. Tl-weighted and T2-weighted scans are MRI scan types that are applied. To point to tissue regions that are alleviated in Tl scans, water and fat molecules are differentiated. Losses of tissues are represented by darker areas. Under the insertion of nonradioactive unit gadolinium, the infected visibility could be enhanced with additional inflammatory lesions. An extraordinary water content tissue is more visible as hyperintense points in T2 scans, which present tissue loss regions.

Cerebrospinal fluid (CSF), white matter (WM), and gray matter (GM) are assumed to be diverse tissues that are present in the brain. Intensity, position, textural features, and tumor structures are more specific whereas MRI brain images develop the segmentation complex. However, it is not sufficient and wavelet features, gray-level- based features, local binary patterns (LBPs), and the Gray-Level Co-occurrence Matrix (GLCM) could be obtained. There is no requirement of labeling; independent tumor segmentation is processed by doctors in disease diagnosis. Among other patients, glioblastoma multiforme (GBM) BTs are commonly detected tumors in the brain. Due to the structure, texture, and shape difference, robotic tumor segmentation is more crucial, which tends to result in false positive, soft tissues, and blood vessels, which are assumed to be nontumor brain structures that are nonidentified as tumors. Various types of conventional models perform completely independent tumor segmentation. Hence, tumor identification is a classical segmentation technique that is manual and not completely automatic.

A numerous sum of brain MRI scans is essential to deploy classifiers like machine learning (ML) for ВТ segmentation with the ground truth of different real-time applications for training. In developing the optimal classification model, factors are assumed to have classification accuracy, algorithm functions, as well as processing resources [8]. With the application of methods like unsupervised classification, such as FCM and Self Organizing Map (SOM), and supervised approaches, such as к-nearest neighbor (K-NN), SVM, and artificial neural network (ANN), the brain MRI is classified. Generative, as well as discriminative approaches, are two types of automated segmentation of ВТ.

For comparison with independent methods, Menze et al.’s [9] previous work represents that models are based on discriminative classification that indicates optimal function. The association between the ground truth as well as input image is known by discriminative models based on feature extraction [10]. With respect to ground truth, the value is valuable and applies the methods of supervised learning in several cases, which acquires a higher data set. To attain the unknown tumor segments, advanced profit in healthy tissues is used. Hence, it is a complex operation to transform previous knowledge into proper probabilistic techniques.

Soft computing models like fuzzy entropy values are used in selecting the best features. GLCM is assumed to be a statistically relied feature extraction method. GLCM distributes the pixel measures in case of identical gray-level values. In the comparison of classification models, the back propagation neural network (BPNN) model has maximum prediction accuracy value. The learning model investigates the application of predefined data to attain training data for extracting Harlick texture features from every MRI [II]. The attained results from existing models reveal that the Elman Network with log sigmoidal activation function is adaptable when compared with alternate ANNs with higher performance value. Under the application of ANN classification, the grades of tumors are computed with a maximum accuracy [12]. The above task applies an automatic prediction as well as the segmentation model. Following, a median filter has been utilized for eliminating gray as well as white noise. The ВТ analysis is forecasted using fast bounding box (FBB) schemes.

The ANN employs features to train data with the help of pretrained data and performs the tumor classification. It applies a grading matrix of Grade (II—IV) of three grades. The histogram is employed to estimate the intensity of 2D images. Also, the obtained histogram signal has been used to Slantlet transform to perform the feature extraction process [5]. Binary classification model is operated according to NN. the system undergoes training under the application of pretrained feature data, and it has been classified automatically using an Alzheimer’s disease pathological brain. Also, various conventional models are used in tumor segmentation. Thus, to identify the tumors, the classical segmentation is manually not accurate. Few w'orks make use of different AI techniques to diagnose the disease effectively [13-17].

This chapter devises an effective IoHT with CC-based ВТ identification model by the use of GLCM and PSO with SVM. The proposed PSO-SVM-based detection model comprises preprocessing, feature extraction, and classification. After completing the preprocessing stage, the feature vectors will be extracted from the preprocessed image. Finally, the PSO-SVM classifier model is applied for the classification of ВТ images as benign or malign. The presented GLCM-PSO-SVM model is validated using a set of images from the brain tumor segmentation (BRATS) data set. The simulation outcome ensured that the GLCM-PSO-SVM model is effective in terms of sensitivity, accuracy, and specificity.

 
Source
< Prev   CONTENTS   Source   Next >