Low Field MRI Deep Learning Framework for Non-Hospitalizing Early Detection and Characterization of Alzheimer’s Disease Pathology

Alzheimer’s Disease (AD) is the most common type of neurodegenerative disease and it significantly disrupts brain function, leading to memory loss, cognitive decline, and more. Currently, there is no effective cure or treatment for AD, making early diagnosis critical so preventative actions can be taken before large-scale degradation occurs. The most common diagnostic techniques involve various neuroimaging modalities, which provide structural and functional information about the brain. However, many of these techniques are extremely expensive, have long scan times, and require access to hospitals equipped with high-grade equipment such as MRI, PET, and CT scanners.To address these challenges, our team focused on the emerging use of Low-Field MRI scanners, which are cheaper, portable, and have much shorter scan times compared to traditional MRI scanners due to their lower magnetic field strength. However, these scanners provide significantly less information than expensive hospital-grade MRI scanners. Therefore, the goal of our study was to develop an image enhancement framework capable of effectively segmenting brain regions associated with AD (such as the amygdala, hippocampus, and ventricles) and automatically diagnosing AD using Low-Field MRI scans.In our approach, we first created a framework to generate synthetic Low-Field (LF) MRI scans using a Fourier Transformation framework. We then developed a deep learning framework to enhance these LF MRI scans by utilizing a SRCNN based architecture and UNET++ models, which performed super-resolution and segmentation, respectively, to obtain volumetric information. Our SRCNN model achieved a Mean Squared Error of 214.54, a Peak Signal-to-Noise Ratio of 31.2 dB, and a Structural Similarity Index Measurement of 0.82. The UNET++ model achieved an accuracy of 96.3 percent, a precision of 89.3 percent, a recall of 85.6 percent, and a Dice score of 0.93.Finally, we utilized the volumetric information to classify the scans as either AD or normal using a soft voting majority framework, where the final diagnosis was based on the majority consensus of models. This soft voting framework consisted of three machine learning (ML) models: Linear Regression, Support Vector Machine, and Multilayer Perceptron. The overall framework achieved an accuracy of 0.96, a precision of 0.96, and a recall of 0.98.