Date of Award


Degree Name

Doctor of Philosophy


Mathematical Sciences


Amy Wagler


Neuroimaging experts in biotech industries can benefit from using cutting-edge artificial intelligence techniques for Alzheimerâ??s disease (AD) and dementia stage prediction, despite the fact that it may be difficult to anticipate the precise stage of AD. AD is one cause of dementia with very limited to no treatment available. Cell-type classification studies are essential for developing novel drugs for this lethal and common disease. Neuronal cell segmentation is the process of identifying and separating individual neurons in an image, typically in order to study their properties or analyze their organization in the nervous system. This is significant because neurological problems and diseases can only be treated effectively when the structure and function of neurons are understood.

Therefore, the goal of this research is to develop a cutting-edge, advanced deep-learning algorithm for addressing this problem. We developed Convolution Neural Networks (CNNs) and Graph Convolution Networks (GCNs) based methods for classifying different AD stages. Four separate models were developed: CNNs built from scratch, VGG-16 with additional convolutional layers, GCNs, and a novel CNN-GCN model for classifying the AD stages. We employed DenseNet, ResNet, MobileNet, InceptionNet, and EfficientNet-B7 transfer learning models trained on ImageNet to classify AD cell types. We then implemented the proposed modified EfficientNet-B7 model for multi-class classification of cell types of AD along with a binary classification of cell types individually. We proposed an image segmentation method based on CNNs and graph attention networks (GATs) for segmenting biomedical images.

We achieved an overall accuracy of 43.83%, 71.17%, 99.06%, and 100% by applying CNNs, VGG16 with additional convolutional layers, GCNs, and the CNN-GCN model, respectively, and the CNN-GCN model showed excellent performance in classifying different stages of dementia. By performing 5-fold cross-validation on the multi-class cell-types dataset, we achieved a training accuracy of 80% and a validation accuracy of 63% for the modified EfficientNet-B7 model. By implementing our proposed U-GAT algorithm for image segmentation, we obtained the highest accuracy of 86.5% and the lowest loss of 0.30 compared to the benchmarking algorithms CNNs, U-Net, and GATs, respectively.

Understanding the stages of AD will assist biotech industry researchers in uncovering molecular markers and pathways connected with each stage. This approach can be used or extended to improve accuracy for detecting the impact of various cell types on novel drug developments in AD.




Recieved from ProQuest

File Size

97 p.

File Format


Rights Holder

Md Easin Hasan

Available for download on Monday, December 01, 2025