An efficient combination of quadruple biomarkers in binary classification using ensemble machine learning technique for early onset of Alzheimer disease

BackgroundAlzheimer’s disease (AD) is a degenerated condition of the brain where memory loss is fully depleted for elderly individual. Efficient machine learning methods are accessible, producing low classification accuracy since single modality features are being evaluated. In this paper, the multimodal approach is developed and execution of comprehensive validation for structural atrophy through Magnetic Resonance Imaging decreases metabolism through Fluorodeoxyglucose Positron Emission Tomography (FDG-PET), and accumulation of amyloid plaques through Pittsburgh compound B (PiB-PET), as well as cognitive assessment for identifying the early onset of AD. It has been stated that additional information from multiple image modalities would ameliorate the classification accuracy while diagnosing early AD. The novel classifier, Adaptive Hyperparameter Tuning Random Forest Ensemble Classifier (HPT-RFE), is proposed for three binary classifications. In this classifier, the tunning of hyperparameters is automated for computing the best features while constructing the optimum size of Random Forest. The advantage of using the classifier is computationally much faster when compared with Support Vector Machine, Naïve Bayes, K-Nearest Neighbour and Artificial Neural Network. Simulation results show that the performance of the Adaptive HPT-RFE classifier has been regarded as best among all binary classifications in the ADNI dataset. For AD versus Normal Control (NC) binary classification, 100% accuracy, sensitivity, and specificity have been achieved, whereas the accuracy of 91% and 100% specificity for NC versus Mild Cognitive Impairment (MCI) classification and 95% accuracy, 100% specificity, 80% sensitivity for AD versus MCI classification are compared with four state-of-the-art techniques.