Turkish Journal of Physiotherapy and Rehabilitation; 32(3) ISSN 2651-4451 | e-ISSN 2651-446X www.turkjphysiotherrehabil.org 4212 AUTOMATED MACHINE LEARNING BASED FUSION MODEL FOR BRAIN INTRACRANIAL HEMORRHAGE DIAGNOSIS AND CLASSIFICATION S.P.Velmurugan 1 , Jenyfal Sampson 2 , S.Diwakaran 3 1,2,3 Assistant Professor, Department of Electronics and Communication Engineering, Kalasalingam Academy of Research and Education, Krishnankoil, India. 1 s.p.velmurugan@klu.ac.in, 2 jenyfal.sampson@klu.ac.in 3 s.divakaran@klu.ac.in ABSTRACT Traumatic brain injury (TBI) remains the main reason for mortality and disabilities. TBI results in intracranial hemorrhages (ICH) and the improper diagnosis of ICH can lead to death. Existing techniques for ICH diagnosis are based on the Computerized Tomography (CT) scans by radiologists for the detection of ICH and locate the regions. But this process is mainly based on the accessibility of professional radiologists. In this view, this paper designs an automated machine learning based fusion model for ICH diagnosis and classification (AMLFM-ICH) model. The proposed AMLFM-ICH model encompasses different processes such as preprocessing, segmentation, feature extraction, and classification. Primarily, Gaussian filtering (GF) technique for removing the noise and smoothen the image. In addition, active contour based segmentation method is implemented to segment the preprocessed image. Besides, a fusion of local binary patterns (LBP) and histogram of gradients (HOG) techniques are employed for feature extraction. Moreover, kernel extreme learning machine (KELM) classifier is implementedforallotting the proper class labels to the applied test images. In order to examine the improved diagnostic performance of the AMLFM-ICH model, a series of simulations take place to highlight the enhanced results over the state of art methods. Keywords: Traumatic brain injury, Intracranial hemorrhages, Machine learning, Image processing, Disease diagnosis I. INTRODUCTION Globally, one of the countries are confronting a significant rise in several medicinal people. This dramatic increase in various limits the patient in accessing the primary doctors/caregivers for better treatment [1, 2]. In this advanced technological era, developing the performance of biomedical and healthcare systems is one of the difficult tasks. Obviously, giving a quality of treatment to the persons with lower cost, in shorter period of time, and addressing the lack of healthcare personnel are the major problems. In previous times, the growth of wearable and IoT devices have enhanced the quality treatment using RPM. Currently, IoT devices role a major part in healthcare applications particularly in diagnosing and detecting various kind of diseases in navigation system [3, 4]. [5] presented an optimized clustering technique for VANET by honey bees and GA modules. Brain hemorrhage is a kind of stroke which is generally affected by an artery in the brain. If no accurate detection and treatment in time sensitive process, then it would lead to lifetime disability/death. The major cause of brain hemorrhage is commonly used high blood pressure, smoking, alcohol, and so on. When heredity is also considered as main factor based on brain hemorrhage. Computerized tomography (CT) scan is explored using radiotherapist for predicting ICH and find infected areas. The medicinal professionals state that when ICH, traumatic brain injury (TBI) takes place which results in mortality or body paralyzes for lifespan when medically fails in precise treatment and diagnosing process [6]. Anupama et al. [7] improve DL based ICH diagnoses by GrabCut based segmentation with SDL, called GCSDL module. The presented technique utilizes Gabor filtering for removing noise, thus the image quality could be increased. Too, GrabCut based segmentation method is employed for identifying the diseased part efficiently in the image. Various conventional and DL methods have been established in this study. As regards the conventional ML techniques. The technique identified the ICH subtypes according to their volume, location, and shape [8, 9]. This method attained 59% specificity and 98% sensitivity for detecting ICH and an average accurateness in