Empirical Assessment of Transfer Learning Strategies for Dementia Classification Using MRI Images

Abstract

Dementia is a debilitating neurodegenerative disorder affecting millions worldwide. Early detection is very crucial for effective management. Magnetic resonance imaging (MRI) offers a noninvasive means to assess structural brain changes associated with dementia. In this study, we propose an empirical evaluation for binary classification of dementia using MRI images, utilizing transfer learning techniques applied to a diverse array of pretrained deep learning models. This paper presents a systematic comparison of the performance of various transfer learning approaches, including feature extraction and fine-tuning, across a spectrum of popular pretrained models, such as visual geometry group (VGG), Inception, ResNet, EfficientNet, and DenseNet. This paper also investigates the effects of the transfer learning approach on classification accuracy. Experimental results show that transfer learning is effective in improving classification performance, and they are validated on a large dataset of MRI scans from subjects with and without dementia. Furthermore, the relative benefits and drawbacks of various transfer learning techniques and pretrained models for dementia classification are revealed by the comparative analysis. The results of this investigation enhance the development of automated diagnostic instruments for dementia, thereby promoting prompt intervention and enhanced patient results.