Applying a deep learning approach focusing on spatiotemporal features in early diagnosis of Alzheimer's disease

BACKGROUND: Alzheimer's disease is a neurodegenerative disorder (Vally & Kathrada, 2019), the most common type of dementia in the elderly population (Kim, 2020) and is a great challenge in the geriatric field. Cognitive impairment is one of the features shown by Alzheimer's dementia patients, and instrumental activities of daily living (IADL) are predictors of cognitive impairment as proven by several research works. Over the past few years, deep learning has had a tremendous improvement effect on diverse science areas, inclusive of healthcare (Farouk & Rady, 2020). This study aims to explore a non-invasive and novel approach with the use of deep learning towards the early diagnosis of Alzheimer's disease. METHOD: To measure cognitive impairment, IADL are objectively recorded, capturing data from egocentric videos using wearable cameras that are attached to a glass frame of each participant, mainly focusing on hands use while performing these activities. Obtained images are analyzed based on human-object interaction and human-environment interaction. To make precise analysis while performing these activities, we propose the use and relationship of the anatomical planes (coronal, sagittal and transverse planes) and healthy human functional patterns [Martínez-Velilla, 2018], in a quantitative way using deep learning. In the coronal plane, patterns involving displacement movements and object manipulation were identified with an accuracy of 87%. The information from the sagittal and transverse planes is developed with a deep learning model, which provides the required depth data to link the IADL's quality. By analyzing these planes, we can get more information about the distance of the hand and body motion while performing these activities. RESULT: From our work, we obtained an accuracy of 87% recognizing movement patterns of displacement and object manipulation, and a good prediction of the depth of the anatomical planes. CONCLUSION: Our model serves as a tool for proactive prediction of Alzheimer's dementia and support in clinical decision-making.