Akan et al. studied the effectiveness of a video-based space-time model by examining structural MRI scans, aiding in the early detection of Alzheimer's disease. By applying MRI slices sequentially like video frames, researchers developed the AlzFormer deep learning framework, which uses a spatiotemporal self-attention model to detect temporal and structural changes in the brain. AlzFormer testing was performed using T1-weighted MRI scans from Alzheimer's Disease Neuroimaging Initiative (ADNI), applying skull stripping and normalization to MNI space as pre-analysis procedures. On the test set, AlzFormer obtained 97% accuracy and balanced F1-scores for each group (AD: 0.94, MCI: 0.99, CN: 0.98). No cases were mistakenly classified as AD, suggesting that dementia patients and healthy individuals have a high level of class separability. AlzFormer consistently outperformed other models in terms of accuracy, precision, recall, and F1-score, even identifying clinically significant brain regions that correlate with patterns of AD progression based on attention map analyses. The researchers state that AlzFormer solves the "black-box" problem in AI by providing improved classification and attention heatmaps for straightforward interpretation. The model does not yet provide disease progression longitudinally and relies on high-resolution MRI, but future improvements will include multimodal and follow-up data.