Various methodologies have been presented for detection and classification of AD over the years, some of them are examined below.
M. Liu et al. [1] created a CNN-based multi-model deep learning framework for learning hippocampal segmentation and illness categorization. It is tested using T1-weighted structural data from the Alzheimer's Disease Neuroimaging (ADNI) database as a baseline. The results, which were based on the ADNI dataset, showed that the technique has a promising performance for diagnosing AD and MCI.
In this study, U. Rajendra Acharya [2] et al. compare T2-weighted brain MRI in people with and without Alzheimer's disease. The aim of this research is to create a Computer-Aided-Brain-Diagnosis (CABD) system that can detect symptoms of Alzheimer's disease in brain scans. For classification, the method employs MRI together with a variety of feature extraction techniques. The results of this study show that the ST + KNN methodology produces better results than other methods. The disadvantage of this study is that the system's performance could be improved by utilizing different classifiers.
For AD classification, I. Beheshti [3] et al. used a Histogram-Based feature extraction (CAD system). There are five steps to it. Pre-processing and a novel statistical feature-generation procedure are among the phases. Statistical information from high-dimensional similarity matrices of varied sizes is compressed into smaller, fixed-size vectors using the proposed feature-generation method. The study's main problem is that, when compared to previous studies, the suggested CAD system's performance in terms of ACC and SPE is extremely competitive.
S. Afzal [4] et al. are interested in applying deep learning algorithms to categorize Alzheimer's patients and normal controls in order to automatically discover biomarkers (hidden cues). Their findings imply that the network can learn to categorize the two extreme classes (NC vs AD) using the existing data, but that when faced with a three-way classification job, it will struggle. The reason for this is not just the scarcity of data, but also its ambiguity: MCI pictures appear to be identical in both classes.
This study by Farheen Ramzan [5] et al. examines the efficiency of rs-fMRI for multi-class classification of Alzheimer's disease. They also recommended using deep residual neural networks in conjunction with a transfer learning strategy to classify the six phases of Alzheimer's disease. For feature extraction and categorization of different phases of Alzheimer's disease, pre-processed data is fed to CNN-based neural networks. The method has shown to be effective in the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
V. Sathiyamoorthy [6] et al, in this study, a novel methodology for predicting AD is provided as a CAD process using several algorithms. ADNI has been used to gather relevant MRI brain pictures. To eliminate noise from the photos, the image restoration technique use the 2D Adaptive Bilateral Filter (2D-ABF) algorithm. The Alzheimer's disease (AD) in MRI brain pictures is automatically segmented and classified. GLCM is utilised to obtain multiple image features, and the DCNN classification approach is employed to classify normal and abnormal images. Using a deep Convolutional Neural Network, this approach achieved high accuracy.
Junhao Wen [7] et al., this research examines seven feature ranking approaches (SD, MI, IG, PCC, TS, FC, and GI) for detecting Alzheimer's disease (AD) using structural magnetic resonance imaging (sMRI) data in order to develop an automatic computer-aided diagnostic (CAD) system. The classification error estimation in the training phase determines the ideal size of top discriminative features. With 10-fold cross-validation, the suggested approach is tested on a data set from ADNI (130 AD and 130HC).
In this study, 1820 T2-weighted brain magnetic images from the ADNI database were used by Shaik Basheera [8] et al. The proposed method involved extracting grey matter from brain voxels and utilising CNN to classify it. They used hybrid improved ICA to construct a deep learning technique for categorization based on the GM segment. They used binary classification techniques including AD-MCI, AD-CN, and MCI-CN, as well as multiclass techniques like AD-MCI and CN.
They also compared the classifier's performance to the physician's conclusion and found that it worked well. Classifying distinct phases of AD, which is a difficult operation due to overlapping traits in different stages of AD, is one of the identified shortcomings of existing methods. Due to limited availability, legitimate datasets are difficult to obtain. Despite the fact that many machine learning models have achieved high accuracy, they have only employed small amounts of data. According to the survey, all preprocessing techniques improve the result, thus this work employs strong noise reduction and image enhancement algorithms, such as CLAHE and TV denoising algorithms, as well as a transfer learning strategy to comprehend and create a new CNN model which provides better results and classifies precisely for large datasets.