Automated Medical Diagnosis of Alzheimer´s Disease Using an Efficient Net Convolutional Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) presents a significant global healthcare challenge.
• Deep learning (DL) has emerged as a key technology for early AD detection since 2017, utilizing neuroimaging biomarkers.
• Convolutional Neural Networks (CNNs) are increasingly applied to analyze complex medical image data.

Purpose of the Study:

• To implement and evaluate an EfficientNet-b0 CNN model for classifying cognitive states in Alzheimer's disease.
• To investigate a novel "fusion of end-to-end and transfer learning" approach for AD classification.
• To assess the model's performance in distinguishing between cognitively normal (CN), stable mild cognitive impairment (sMCI), and Alzheimer's disease (AD) subjects.

Main Methods:

• Utilized 245 T1W MRI scans from CN subjects, 229 from AD subjects, and 229 from sMCI subjects.
• Applied a standard preprocessing pipeline to all MRI scans.
• Trained and evaluated an EfficientNet-b0 CNN model incorporating a hybrid learning strategy.

Main Results:

• For sMCI vs. AD classification, achieved 93.10% accuracy and 93.00% AUC on test data.
• For multiclass AD vs. CN vs. sMCI classification, achieved 87.38% accuracy and 88.00% AUC on test data.
• Demonstrated high performance in both binary and multiclass classification tasks.

Conclusions:

• CNN-based DL models are effective for analyzing intricate MRI features in clinical settings.
• The proposed fusion of end-to-end and transfer learning enhances AD classification accuracy.
• This approach holds potential for early and accurate diagnosis of Alzheimer's disease stages.