Collaboration

Paper 1: Classification of Alzheimer's Disease using Ensemble of Deep Neural Networks Trained Through Transfer Learning

Publisher:  IEEE Journal of Biomedical and Health Informatics (IEEE JBHI), 2021. (Impact Factor 2020: 5.772).

Abstract:  Alzheimer’s disease (AD) is one of the deadliest neurodegenerative diseases ailing the elderly population all over the world. Many researchers are using deep learning (DL) techniques to learn highly complicated patterns from MRI scans for the detection of AD. It is also found that an ensemble of predictions from multiple models gives better performance as compared to that of a single model. Two major bottlenecks for developing ensemble of DL models are their high computational complexity and requirement of large sample size for better generalization. In this work, we deal with the aforementioned bottlenecks and propose a computationally efficient, DL-architecture agnostic, ensemble of deep neural networks named ‘Deep Transfer Ensemble (DTE)’ trained using transfer learning for the classification of AD. The proposed ensemble leverages the diversity introduced by many different locally optimum solutions reached by individual networks through the randomization of hyper-parameters. The proposed ensemble model also introduces further diverse predictions by exploiting complementary feature views. We also test the model vigorously by analyzing its performance on a large and a small dataset downloaded from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) archive. The DTE utilizes the advantages of random search, transfer learning, and snapshot ensembles in a single ensemble to produce better generalization performance. DTE achieves an accuracy of 99.05% and 85.27% on two independent splits of the large dataset for cognitively normal (NC) vs AD classification task. For the task of mild cognitive impairment (MCI) vs AD classification, DTE achieves 98.71% and 83.11% respectively on the two independent splits. DTE also performed reasonably well on a small dataset consisting of only 50 samples per class. It achieved a maximum accuracy of 85% for NC vs AD on the small dataset. DTE outperformed snapshot ensembles along with several other existing deep models from similar kind of previous works by other researchers.