IoHT-Based Improved Grey Optimization with Support Vector Machine for Gastrointestinal Hemorrhage Detection and Diagnosis Model


Advanced development of the Internet of Health Things (IoHT) has seen a considerable shift on the way to healthcare technology. IoHT is helpful in the healthcare sector, and is useful for clinics as well as homes. Therefore, there is a requirement for scalable intelligent algorithms, which have resulted in high interoperable solutions and decision-making in IoHT. Wired endoscopy methods are extremely applied for diagnosing and observing the anomalous in the gastrointestinal (GI) tract, such as obscure GI bleeding, Crohn’s disease, cancer, and celiac disease. Although it is efficient and stable, conventional endoscopy might cause uneasiness and establish complexities in patients since a longer and flexible tube has to be provided within the GI tract [I]. Additionally, it is tedious to observe the predefined regions of the GI tract, such as larger portions of the small intestine. In addition, the endoscopes require well-trained experts for handling the devices, but also require a longer duration [2]. Finally, technological development as well as effective medical illustrations lead to entirely noninvasive endoscopic systems that do not require sedation, and are available for diagnosing diverse GI malfunctions.

A common Wireless Capsule Endoscopy (WCE) approach is comprised in Fig. 5.1. Alternate capsules apply various sensors, namely, a temperature sensor, pH sensor, and pressure sensor, to evaluate diverse physiological parameters. However, the capsule endoscopy system gains more attention and demonstrates efficiency, but is filled with shortcomings. Some of the constraints are minimized battery-life, poor image quality, absence of localization, and dynamic locomotion controlling.

The task of classifying bleeding and nonbleeding images from WCE images deals with various complications. The battery power of attained final outcome limits the

General structure of the WCE approach

FIGURE 5.1 General structure of the WCE approach.

outcome when it has a minimum resolution of captured frames. Simultaneously, it provides a slower frame value of 2 frames/s. Also, about 6,000 images undergo examination for each iteration. This investigation helps the physician to provide 120 minutes for analyzing an image, which is not applicable to real-time scenarios. Since the validation process consumes a higher duration, the exploring process of bleeding is prone to human fault. Therefore, the automatic prediction method for bleeding frames develops numerous operations for doctors. Suspected Blood Indicator (SBI) [3] is applied to perform the automatic prediction of bleeding frames. However, SBI shows minimum sensitivity and specificity and constantly failed in finding different types of bleeding of the small intestine.

A program deployed with Given Imaging Ltd. allows the physician to look at a couple of successive frames at the same time. Therefore, due to the presence of a minimum rate of the frame, a couple of successive frames do not acquire the concerned area. Consequently, the physician should snap between the images to perform the verification process, which is overburdened and consumes maximum duration. Hence, automatic detecting approaches solve the limitations involved in this model. The primary schemes of GI hemorrhage identification are classified into color as well as texture, and both color through texture-centric methods. The predetermined techniques [4] majorly exploit the ratio of intensity rates in red green blue (RGB) or Hue, Saturation, Index (HSI) domain. The second approach focuses on applying textural data of bleeding as well as nonbleeding images to perform the classification process [5]. It can be clear that a combination of color and texture descriptors provides optimal outcome in terms of accuracy.

The techniques from the initial variety are more rapid, but they do not process the task of finding tiny bleeding portions. The pixel-centric model works on each pixel of an image to produce the feature vectors. Finally, it is comprised of higher processing complexity. The third patch provides maximum sensitivity by giving specificity as well as accuracy. Additionally, the data patch has to be predicted in a manual fashion that conceals the method producing the whole process in an automatic manner. Li and Meng [6] implied a chrominance moment as well as the Uniform Local Binary Pattern (ULBP) approach for detecting the part of bleeding. Hwang et al. [7] signified a super-pixel and red ratio-relied solution that offers good function. Therefore, a high processing cost is observed and it has failed to perform effectively on the images with minimum illumination and better angio- dysplasia region.

Few researchers apply MPEG-7 depending on the visual descriptor to find the medical actions [5]. Pan et al. [8] projected a 6-D feature vector used in the probabilistic neural network (PNN) as the classifier. Liu and Yuan [9] presented Raw, Ratio, as well as Histogram feature vectors, which are basically the intensity measures of pixels, and utilized SVM for detecting GI bleeding images. Hegenbart et al. [10] exploited scale-invariant wavelet-based texture features to predict Celiac disease through endoscopic videos. Under the application of MPEG-7 depending on the visual descriptor, a Bayesian and SVM, Cunha et al. [11] segmented the GI tract into four major topographical regions and classified the images. Some other medical diagnosis models were developed in [11-16].

This chapter presents a new Improved Grey Wolf Optimization (IGWO)-based support vector machine (SVM) called the IGWO-SVM model for the detection of bleeding regions from WCE images. The proposed method contains the group of different processes namely data collection, preprocessing, feature extraction, and classification. Once the data is collected and preprocessed, a proficient normalized gray-level co-occurrence matrix (NGLCM) technique is utilized to extract the features from the provided GI images. After that, the classification process is carried out by the use of NGLCM-IGWO-SVM, where the parameters of SVM have been tuned by the IGWO algorithm. The simulation of the NGLCM-IGWO-SVM model takes place using benchmark GI images.

< Prev   CONTENTS   Source   Next >