Predicting the Severity of Alzheimer’s/Dementia from Magnetic Resonance Imaging Scans Using a Deep Learning Approach

The overall approach used in this research was utilizing a deep learning model, specifically a convolutional neural network. A convolutional neural network is a type of neural network architecture that specializes in identifying and making sense of patterns in image data. For this research, this type of neural network was mostly used for classifying images of dementia MRI scans based on their severities.

Dementia and Alzheimer’s disease are growing health challenges, as an estimated 10 million new diagnoses happen annually, highlighting the need for more efficient diagnostic processes. The lengthy process involves scheduling appointments, pre-screening assessments, and arranging screening appointments, consuming valuable time for healthcare professionals. A streamlined process leveraging computer-based screening could optimize efficiency. In fact, artificial intelligence-based systems also have potential to decrease overall healthcare costs by upwards of 10%, or over $360 billion annually (The Financial Cost of AI in Healthcare, 2023). Deep learning models, with their precision and technical capabilities, offer a promising solution. By automating screening tasks using deep learning, they can reduce administrative burdens, enhance accuracy, and expedite diagnoses, ultimately improving patient care and outcomes. The overall approach used in this research was utilizing a deep learning model, specifically a convolutional neural network. A convolutional neural network is a type of neural network architecture that specializes in identifying and making sense of patterns in image data. For this research, this type of neural network was mostly used for classifying images of dementia MRI scans based on their severities. It used certain classifying methods such as multi-class classification and a high volume of data (roughly 6400 images) to accurately distinguish between different levels of dementia severity. The model's performance metrics are robust, achieving an accuracy of 0.989, a precision of 0.99, a recall of 0.989, and an f1-score of 0.99 after 20 training epochs. These results suggest a high potential for CNN-based models to be integrated into clinical workflows, providing rapid and reliable assessments of dementia severity.

Explore More!

Research Paper