Neurocognitive Impairment Detection in The Elderly: A Holistic Deep Learning Approach

Neurocognitive disorders are becoming more prevalent as the global population ages, making early and accurate detection increasingly important. Traditional diagnostic methods often focus solely on clinical assessments or imaging techniques, which may not give an entire view of a person’s cognitive health. This study introduces a holistic deep learning approach that integrates non-invasive patient data with MRI scans to enhance diagnostic accuracy. Specific Deep Learning algorithms like TabNet with prior Exploratory Data Analysis is used to analyze structured clinical data, while EfficientNetB0 algorithm processes MRI images of both non-demented and mildly demented brains, aiding in early detection. A late fusion strategy combines insights from both models, a multimodal learning approach, leading to improved predictive performance. This approach is designed to be efficient for all systems, making it accessible for real-world applications. The experimental results show that integrating multiple data sources leads to more reliable detection of early neurocognitive impairment, which will be very helpful to the healthcare professionals to make better decisions.