Toward accurate Alzheimer's detection: transfer learning with ResNet50 for MRI-based diagnosis

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) is a leading cause of dementia, affecting over 50 million globally.
• Current diagnostic methods for AD using MRI often require manual feature extraction, which is time-consuming and limits scalability.
• The projected tripling of AD cases by 2050 necessitates automated, accurate diagnostic tools.

Purpose of the Study:

• To develop and evaluate an automated feature-extraction approach for Alzheimer's disease (AD) detection using brain MRI scans.
• To compare the performance of deep learning-extracted features classified by Softmax, Support Vector Machine (SVM), and Random Forest (RF) algorithms.
• To assess the efficacy of transfer learning with a pre-trained ResNet50 convolutional neural network (CNN) for AD diagnosis.

Main Methods:

• Utilized a pre-trained ResNet50 CNN for automated deep feature extraction from brain MRI scans.
• Classified extracted features using Softmax, SVM, and RF algorithms.
• Evaluated model performance on two publicly available datasets: Alzheimer's Disease Neuroimaging Initiative (ADNI) and MIRIAD.

Main Results:

• The ResNet50-CNN combined with the Softmax classifier achieved superior performance, exceeding 99% accuracy, sensitivity, and specificity on the ADNI dataset.
• This model surpassed existing state-of-the-art benchmarks for AD detection.
• The ResNet50-Softmax model demonstrated strong performance on the MIRIAD dataset, achieving 96% accuracy.

Conclusions:

• Transfer learning with ResNet50 significantly improves the accuracy and scalability of Alzheimer's disease diagnosis from MRI data.
• The automated approach eliminates manual feature engineering, streamlining clinical neuroimaging workflows.
• Deep learning models show great promise for supporting early AD diagnosis and addressing the growing global health challenge.