Transfer Learning-Based Ensemble of Deep Neural Architectures for Alzheimer's and Parkinson's Disease Classification

The use of transfer learning in medical imaging has shown promising results in various applications, including disease classification and segmentation. Early detection of neurological diseases like Alzheimer's (AD) and Parkinsons (PD) is the need of the hour. This research experiments MRI datasets pertaining to AD and PD using transfer architecture of neural networks for disease classification. We used three popular datasets, namely ADNI, OASIS, and NTUA, and evaluated seven state-of-the-art transfer learning algorithms for classification. The experiments demonstrates the effectiveness of transfer learning in Alzheimer's and Parkinson's disease classification by achieving high accuracy and AUC scores. While the study highlights the top performing neural network models like InceptionV3 and InceptionResNetV2 for both OASIS and ADNI, it also showcase the high performances of transfer architectures like ResNet50 and EfficientNetB0 from the NTUA dataset. Additionally, we presented an ensemble of these algorithms. Relevant codes can be found at https://github.com/snoushath/AD- PD- TransferLearning.git