Optimized CNN Architecture for Early Detection of Mild Cognitive Impairment

Mild Cognitive Impairment (MCI) is an early stage of Alzheimer’s disease that increases the risk of dementia if not diagnosed as early as possible. Early detection is crucial, as timely interventions can help slow its progression. Traditional diagnostic methods rely on neuropsychological assessments, which are subjective, and biomarker-based tests, which are expensive. Convolutional Neural Networks (CNNs) proves to be an unavoidable option for classification of MCI classification using brain MRI scans. This work uses the ADNI dataset, containing 247 images for Cognitive Normal (CN) and 410 images for MCI. We have deployed three pre-trained CNN models such as MobileNet, EfficientNet, and ResNet, for transfer learning. Since the classification accuracies of these pre-trained models are not up to the expectations, a custom-designed architecture is proposed, which outperforms the pretrained models, giving $99.81 \%$ classification accuracy. The proposed customized CNN model enhances classification accuracy and supports early diagnosis. The applications of this work include early diagnosis of Alzheimer’s disease, clinical decision support, personalized treatment plans, research and development, telemedicine and remote monitoring, educational tools, and public health initiatives.