Skip to main content
Fig. 2 | Alzheimer's Research & Therapy

Fig. 2

From: An explainable self-attention deep neural network for detecting mild cognitive impairment using multi-input digital drawing tasks

Fig. 2

Visual explanations provided by the multi-input VGG16 model with Grad-CAM visualization (2nd column from the right) and the proposed model (column on the far-right) on a representative MCI test sample (2nd column from the left). For the clock image (1st row), our model highlights the hands of the clock where it says 12:55 instead of 11:10. For the cube-copying image (2nd row), our model highlights unusual paths better. For the trail-making test (last row), our model could focus along the paths that should not have been drawn (paths from 2-3, B-4 and C-D), while the multi-input VGG16 model with Grad-CAM failed to highlight some of those paths (B-4). Note that the red arrow and asterisks were not drawn by the subjects but added here to aid the descriptions

Back to article page