Abstract
Dementia, a chronic neurodegenerative disorder, progressively impairs cognitive functions such as memory, reasoning, learning, and recall, placing a significant burden on patients and healthcare systems. Early and accurate classification of dementia severity is crucial for personalized care and intervention. This study introduces a novel Convolutional Neural Network (CNN) designed to classify dementia into four ordinal severity levels (None, Very Mild, Mild, and Moderate) based on MRI brain scans. Utilizing the extensive Open Access Series of Imaging Studies (OASIS) dataset, which includes 86,437 MRI scans (67,222 ‘none,’ 13,725 ‘very mild,’ 5,002 ‘mild,’ and 488 ‘moderate’), our model addresses severe class imbalance with a combination of the Synthetic Minority Over-sampling Technique (SMOTE) and advanced over- and under-sampling methods. By implementing ordinal classification, the model effectively captures the progressive nature of dementia, showing comparable or improved performance against current diagnostic benchmarks. This approach highlights the benefits of ordinal classification in medical imaging, paving the way for enhanced severity assessment and supporting better treatment planning.
Recommended Citation
Bhadrashetty, Ambresh and Sandhya, P.
(2024)
"Convolutional Neural Networks for Dementia Severity Classification: Ordinal Versus Regular Methods,"
Iraqi Journal for Computer Science and Mathematics: Vol. 5:
Iss.
4, Article 14.
DOI: https://doi.org/10.52866/2788-7421.1224
Available at:
https://ijcsm.researchcommons.org/ijcsm/vol5/iss4/14
