On Improved 3D-CNN-Based Binary and Multiclass Classification of Alzheimer's Disease Using Neuroimaging Modalities...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder affecting elderly populations globally.
• Neuroimaging techniques like MRI and PET are crucial for detecting pathological brain changes in early AD.
• Deep learning (DL) models, particularly Convolutional Neural Networks (CNNs), show promise in analyzing medical images for disease diagnosis.

Purpose of the Study:

• To evaluate the impact of different data augmentation techniques on CNN performance for early Alzheimer's disease diagnosis.
• To compare the effectiveness of various augmentation strategies in 3D neuroimaging (MRI and PET) classification tasks.
• To investigate the influence of data augmentation versus architecture design on diagnostic accuracy.

Main Methods:

• Three distinct data augmentation techniques were applied to 3D MRI and PET neuroimaging datasets.
• Convolutional Neural Network (CNN) architectures were employed for both binary and multiclass classification of Alzheimer's disease.
• Performance metrics were analyzed to compare the efficacy of each augmentation method and architectural variations.

Main Results:

• Random zoomed in/out augmentation demonstrated superior performance compared to other tested methods.
• Combining multiple data augmentation techniques often led to decreased classification performance.
• Data augmentation strategies had a more significant impact on performance than architectural modifications or network depth.

Conclusions:

• Random zoomed augmentation is a highly effective technique for enhancing CNN-based early Alzheimer's diagnosis using neuroimaging data.
• Careful selection and application of data augmentation are critical, as inappropriate combinations can be detrimental.
• Focusing on data manipulation techniques may yield better diagnostic improvements than solely relying on complex network architectures or deeper models.