A Hybrid Siamese-LSTM (Long Short-Term Memory) for Classification of Alzheimer's Disease

Introduction

In order to save human lives and improve healthy living, the medical issues have become the most popular problem in the world. Both the researchers and medical experts are working tirelessly at medical diagnosis treatments, advancements and examinations. One of the main medical issues that challenges human health is Alzheimer ’s disease (Bi & Wang, 2019; Choi, Kim, Yoon et al, 2020; Duc et al., 2020). The common brain diseases are vascular dementia and (AD). It is a type of gradual and neurodegenerative disease. Although, the disease is incurable, the early diagnosis of the disease can help the disease from deteriorating. At present, the method that is utilized for Alzheimer’s Disease is testing the (MRIs), to get a result with respect to the disease. Because of the characteristics of high resolution analysing the MRIs can assist differentiating between AD and Normal (Lee et al., 2019; Qiu et al., 2020; Raju et al., 2020). Nevertheless, the method is time-consuming and it may lead to misdiagnosis and since the symptoms varies among person. Early examination of Alzheimer’s is basically affiliated with the AD prodromal level identification, (MCI) Mild Cognitive Impairment. MCI is noted to have a powerful chance of advancement to Alzheimer Disease. (Acharya et al., 2019; Gautam & Sharma, 2020; Liu et al., 2018).

In medical images like MRI, X-ray, CT, mammography and microscopy the deep learning techniques have achieved the major victory. Recently, the MRI scans used in imaging of the brain provides information about the white matter, gray matter, and CSF’s structure (Puente-Castro et al., 2020; Sivaranjini & Sujatha, 2020). In the presented model by using the algorithm of K-means clustering the initial stage of tumor detection can be obtained automatically. The MRI system generates the samples of the brain, i.e., the machine learning will detect various portions in the brain (Oh et al., 2019; Qiao et al., 2018). Firstly, the brain images were gathered from the dataset (ADNI). By practicing histogram equalization, the unwanted noises in the collected brain images were removed (Son et al., 2020; Sun et al., 2020; Tufail et al., 2020). For skull removal, the pre-processed brain images were utilized. For segmenting the white matter, CSF and grey matter after the skull removal is done, segmentation was carried out using FCM (Choi, Kim, Yoon et al, 2020; Mendoza-Léon et al., 2020).

In the prior works, a Siamese CNN, which concatenates identification and verification models, is adopted for classification of remote sensing scene, a Siamese-CNN inspired by VGG-16 to classify the stages of dementia, combining methods of three feature extraction (LBP, HOG and GLCM) with relief feature selection in order to enhance the Alzheimer disease classification, for classification an extensive brain tumor the segmentation and classification the VGG19, CNN model is used on Magnetic Resonance Images data techniques were used. Nevertheless, the related machine learning and DL techniques were trained however have few restrictions (Altaf et al., 2018; Choi, Madusanka, Choi et al, 2020).