Using CNN Saliency Maps and EEG Modulation Spectra for Improved and More Interpretable Machine Learning-Based Alzheimer’s Disease Diagnosis

<div>Steps to perform database partitioning to avoid information leakage. (a) First, 460 segments per subject are split into 5 parts of 92 segments. The last seven are discarded to avoid data leakage. (b) Disjoint parts are then combined per subject. (c) Lastly, temporal shuffle is done to avoid ordering effects. 1/5, 1/5, and 3/5 of the shuffled data are partitioned into validation, testing, and training subsets.</div>

Computational Intelligence and Neuroscience

fig4

Figure 4

Figure 4: Using CNN Saliency Maps and EEG Modulation Spectra for Improved and More Interpretable Machine Learning-Based Alzheimer’s Disease Diagnosis 

Figure 4 | Using CNN Saliency Maps and EEG Modulation Spectra for Improved and More Interpretable Machine Learning-Based Alzheimer’s Disease Diagnosis 