Alzheimer’s disease (AD) is a progressive neurodegenerative disease that is irreversible. Alzheimer's disease causes the brain to shrink (atrophy) and brain cells to die. According to WHO, Alzheimer’s disease is the 7th deadliest disease in the world. The main cause of the disease is the formation of tangles and plaques in the brain, that obstructs the communication between neurons. Symptoms of the disease include a decline in thinking, behavioral and social skills that affect a person's ability to function independently. There is no treatment that cures Alzheimer's disease or alters the disease process in the brain. In advanced stages of the disease, complications from severe loss of brain function such as dehydration, malnutrition, or infection and result in death.
The problem - Our Motivation
The brain is the primary organ of the human body. The diseases that affect the brain are very crucial to handle since most of the changes that occur in our brain are irreversible in extreme cases. Studies show that humans with Alzheimer’s disease exhibit symptoms at the age of their mid-60s with the early onset occurring anywhere between their 40s and 50s. Scientists have not yet found a pharmacological treatment to stop the progressive nature of the disease. According to World Health Organization (WHO), by 2025, the aging population of 65 and above predicted to be 800 million people among them two thirds will be from developing countries. Alzheimer’s disease is a public health issue that is both broad and intimate in scope. Without action it can only be portend to worsen. According to ADI, there are some drugs that can slow down the progression in some patients with Alzheimer’s for a period between 6 and 18 months. Thus, giving us hope that a novel solution in early detection would help in delaying the onset of AD.
What we aimed for
To detect Alzheimer’s Disease from Magnetic Resonance Imaging (MRI). To separate the region of interest (RoI) that are influencing the decision of deciding the AD efficiently using deep learning techniques. To design an AI framework to detect the stage of AD. To make the outcome easily interpretable for the clinicians and doctors.
Structural analysis of brain MRI is done to detect the abnormalities in the brain. The dataset requires for this study is obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI). There are two types of data that are used for classification. One is T1-weighted MRI data in .nii format which is Structural MRI and the other is Cerebrospinal fluid (CSF) data.
The task in front of us
Alzheimer’s Disease is mainly divided into three stages. First stage Normal Cognition (NC) where the patient’s brain functionality is normal. The second stage is Mild Cognitive Impairment (MCI), in this stage dementia is observed in the patients, as MCI progresses it leads to Alzheimer’s disease (AD) which cannot be cured. Our model provides significant improvement for multiclass classification. The three-stage classification of AD is presented below in the form of images.
Cerebrospinal fluid (CSF) is a clear, colorless body fluid found within the tissue that surrounds the brain and spinal cord of all vertebrates. CSF is in direct contact with the brain's extracellular space and thus should reflect ongoing biochemical changes occurring in the central nervous system (CNS), providing a potential window for AD-related changes in the brain. CSF markers with the volumes of the hippocampus can be combined for distinguishing between NC, MCI, and AD. The CSF has proteins namely, Abeta42, Abeta40, total tau (T-tau), and phosphorylated tau (P-tau) As Alzheimer’s disease progresses: 1) Ratio of ABeta42 to ABeta40 decreases 2) P-Tau increases T-tau increases and 3) Aβ42 decreases Ratio of P-Tau to ABeta42 increases.
Segmentation- Our first step
Segmentation is the extraction of the required region of interest(RoI) from the MRI scans. For segmentation of the Brain MRI for the hippocampal region, we have used a deep 3-D convolutional network. From the segmented images, the volume of the hippocampal region can be calculated. This can give the doctors better and accurate calculations.
Challenges we ran into
We worked on all the experiments using Google Colab, so the major challenge that we faced was GPU limitations. It took us a lot of time to run even a single epoch causing time limitations. The dataset was also a difficult one to obtain for the CSF biomarkers.
Accomplishments that we're proud of
Segmentation of the hippocampus area and quantification of the hippocampal region are done with the 3D-CONV net using Brain MRI. The classification of NC and MCI is done first using the hippocampus volumes. Then these MCI subjects are further classified as AD and MCI using the CSF biomarkers. The classification of different stages of AD is done using two different biomarkers. This method of using different features to classify separately is the innovation here. Using this innovative approach, we were able to get much more accurate results.
We reduced our false positives and negatives as much as possible.
What's next for Early on-set detection of Alzheimer's disease from MRI scan
In the future, we can classify the stage of the AD of the patient and predict the timeline of the advancement of the disease by loading the MRI data of the patient on a web page. The backend of the webpage processes the data and finds the hippocampal volume of the patient. Using the hippocampal value, the AI model will output the stage of AD and the time period of progression to the final stage of AD. This helps in predicting suitable medicine to delay the on-set of AD.