Light Weight Approach for Real time dementia Classification Using Machine Learning

For the improved patient management and timely clinical intervention, detecting the dementia in the early stage plays as a vital role. This paper gives a lightweight deep learning study approach for dementia classification. Working the MobileNetV2 architecture, this suits properly in real-time deployment and in the environments which are resourceconstrained. Experiments were conducted on the publicly available Kaggle Alzheimer’s MRI dataset. $\mathbf{6, 4 0 0}$ brain MRI images divided into four classes for the training and evaluation in this study. Specific planned model was finetuned using transfer learning, skilled with categorical crossentropy loss function, then working was improved with the Adam optimizer. The experimental results show that MobileNetV2 achieves an overall accuracy of 89%, with a higher recall value of 99.5% to the Non-Demented style, by representing solid sensitivity in distinguishing healthy subjects from those with mental fall. Subsequently, the aim of this work is to expand a lightweight deep learning study model constructed on MobileNetV2 for classifying dementia stages by working on grayscale MRI images. Specific planned approach focuses on achieving high accuracy with low computational cost, permissive real-time and resourceefficient deployment in clinical and remote healthcare settings. Future work will put on addressing these limitations through state-of-the-art enhancement, class- weighted loss, and attention system to help sensitivity for early dementia detection. Specific findings feature the ability of MobileNetV2 as a computationally active framework for dementia screening applications.