Abstract

Accurate multi-stage classification is necessary for early diagnosis and effective treatment of Alzheimer's disease (AD). In this study, we utilize brain MRI data from the OASIS dataset to classify AD into four stages—Non-Demented (ND), Very Mild Dementia (VMD), Mild Dementia (MD), and Moderate Dementia (MoD). We evaluate two convolutional neural network architectures—VGG19 and ConvNeXt V2—on this task. VGG19 has a slightly higher accuracy (95.47%) and F1-score (0.95) than ConvNeXt V2 (94.06% and 0.94), but both models detect MoD cases with 100% precision and recall. We used Local Interpretable Model-Agnostic Explanations (LIME) and Randomized Input Sampling for Explanation (RISE) to make the models easier to understand after use. The differences between how well the AI explains its decisions and how accurately it diagnoses Alzheimer's disease are shown by a detailed comparison of transformer-based models and traditional CNNs.

Concepts :

Access to Document

SDGs

Citations by Year