Abstract:
"The World Health Organization (WHO) has identified Alzheimer’s disease as a worldwide health priority. Early detection of the disease is critical since the traditional disease identification methods take a long time to detect the disease. There are many machine learning techniques proposed for detecting Alzheimer’s disease at an early stage. The traditional approaches show low detection accuracy and performance due to the high complex indicators and characteristics of the disease.
The NeuraSage project addresses the identified gap by offering an innovative approach to improve the accuracy and performance of Alzheimer's disease detection, which uses both brain MRI images and textual MRI data. This multimodal approach uses the features from both brain MRI images and textual MRI data to identify early sign and stage of Alzheimer's disease more successfully than traditional single input methods.
The proposed multimodal deep learning approach consists of an ensemble learning approach for the image classification task, which employs two pretrained image classification algorithms (Transfer Learning), while the textual classification is also conducted through an ensemble learning approach employing two text classification models. Then, the two model predictions are concatenated using a late data fusion approach, which would increase the detection accuracy and the model performance.
After comprehensive model testing and evaluation, the proposed ensemble learning approach for the image classification task achieved an accuracy of 99.70%, while the text classification ensemble learning approach achieved an accuracy of 92%. Considering the model performances above, it suggests that concatenating the predictions of these models through a data fusion approach would enhance the detection accuracy of Alzheimer’s disease."
