Biomarkers

BACKGROUND: The imbalance in medical imaging datasets and concerns over patient privacy present challenges in AI-driven dementia diagnosis. This study leverages synthetic data generated by Denoising Diffusion Models (DDM) to address these issues, marking a significant advancement over traditional GAN-based augmentation techniques. METHOD: Using the Kaggle Alzheimer's MRI dataset as a base, this research pioneers the integration of DDM with a novel Conditional Deep Convolutional Neural Network (C-DCNN) for classifying dementia stages. The dataset comprises axial MRIs of four categories: No Impairment, Very Mild Impairment, Mild Impairment, and Moderate Impairment, balanced with 2,560 synthetic images per class. Preprocessing included skull removal, grayscale conversion, and resizing to 128×128 pixels. The DDM-generated synthetic images were validated by a radiologist to ensure clinical relevance. RESULT: The proposed C-DCNN model achieved a remarkable accuracy of 98%, significantly outperforming established architectures like ResNet50, VGG16, VGG19, and InceptionV3. CONCLUSION: This study demonstrates the efficacy of DDM-generated synthetic datasets in improving dementia classification accuracy, setting a new benchmark for AI applications in healthcare. The proposed framework offers a scalable and robust solution for early diagnosis and treatment planning in dementia care.