Affiliations: [a] Department of Industrial Engineering, College of Garmsar, Amirkabir University of Technology (Tehran Polytechnic), Garmsar Campus, Tehran, Iran | [b] Department of Industrial Engineering and Management Systems, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Tehran, Iran | [c] Department of Industrial Engineering, Amirkabir University of Technology (Tehran Polytechnic), Garmsar Campus, Iran
Correspondence:
[*]
Corresponding author: A. Seifi, Department of Industrial Engineering and Management Systems, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran. E-mail: aseifi@aut.ac.ir.
Abstract: The significance of data analytics and machine learning has increased in managing vast quantities of healthcare data effectively. According to recent statistics, Alzheimer’s disease is the most prevalent cause of dementia worldwide. Moreover, Alzheimer’s arises in some people with increasing mild cognitive impairment (MCI). Early detection and treatment of Alzheimer’s disease and its prodromal stage, also known as MCI, is vital to prevent its progression. For selecting the most relevant features, the four feature selection methods, including Mutual Information (MI), Recursive Feature Elimination (RFE), T-test, and Genetic algorithm (GA), are used. Also, three classification methods include Adaboost, random forest, and support vector machine (SVM) with the linear kernel; Moreover, a resnet network is applied to categorize the patients as cognitive normal, MCI, which includes both progressive (pMCI) and stable (sMCI) cases, and Alzheimer’s disease using positron emission tomography (PET) scan images. Among these machine learning methods, combining a t-test and a genetic algorithm for selecting the most relevant features and applying a support vector machine with 8-fold cross-validation have produced the best results on high-dimensional images of Alzheimer’s disease neuroimaging initiative (ADNI). The proposed method differentiates between sMCI and pMCI patients with a 95.45% accuracy rate and 95.23% F1-score, outperforms the performance of recent studies, as well as AD and CN with 97.36% accuracy rate and 100% recall, making it acceptable for supporting clinical applications.
Keywords: Alzheimer’s disease, healthcare system, PET images, classification, machine learning
