Table 1 Features extraction from deep models and comparison of internal validation results with external test result.
From: Multimodal deep learning models for early detection of Alzheimer’s disease stage
Models | Biomarkers extracted | Internal cross validation performance | External test performance | |
|---|---|---|---|---|
EHR (deep models) (CN, MCI, AD) | Regularization coefficients (0.03, 0.03) Dropouts (0.6, 0.6, 0.6) Layer sizes (200, 100, 75) | Memory summary score RAVLT memory test (learning) RAVLT memory test (learning) baseline Neurophysiological battery (AVTOT 6 trials) Metabolomics marker (pe.P.16.0 22.6) | Accuracy: 0.78 ± 0.03 Precision: 0.78 ± 0.04 Recall: 0.78 ± 0.05 F1 Scores: 0.77 ± 0.04 | Accuracy: 0.76 Precision: 0.76 Recall: 0.77 F1 Scores: 0.76 |
Imaging (deep models) Prediction (CN, AD) | Highest on validation (Dropout-0.5, Batch size 5 , Layer size(20), # areas = 5) Highest on external test (SVM kernel = linear) | Left hippocampus Right hippocampus Right superior temporal Right amygdala Left amygdala | Accuracy: 0.86 ± 0.04 Precision: 0.86 ± 0.04 Recall: 0.87 ± 0.04 F1 Scores: 0.86 ± 0.04 | Accuracy: 0.84 Precision: 0.83 Recall: 0.83 F1 Scores: 0.83 |
SNP (deep models) Prediction (CN, MCI/AD) | Regularization coefficients (0.03, 0.03), Dropouts (0.6, 0.6, 0.6) Layer sizes (200, 100, 50) | Gene1 location 207782707 Gene1 location 55342929 Gene10 location 106979076 Gene10 location 50858045 Gene11 location 121493001 | Accuracy: 0.89 ± 0.03 Precision: 0.9 ± 0.04 Recall: 0.84 ± 0.03 F1 Scores: 0.86 ± 0.04 | Accuracy: 0.66 Precision: 0.66 Recall: 0.57 F1 Scores: 0.53 |
EHR + SNP + Imaging (deep models) Prediction (CN, MCI, AD) | Regularization coefficients (0.03, 0.03) Dropouts (0.6, 0.6, 0.6) Layer sizes (200, 100, 50) Random Forest Trees = 31 | Voxel based morphometry Angular left Biomarker (PtdCho 16:0/18:1) MR volumes posterior limb of internal capsule including cerebral peduncle right Biomarker (PC ae C40:5) Biomarker (PC ae C42:4) | Accuracy: 0.79 ± 0 Precision: 0.79 ± 0.07 Recall: 0.79 ± 0.07 F1 Scores: 0.79 ± 0.07 | Accuracy: 0.78 Precision: 0.77 Recall: 0.78 F1 Scores: 0.78 |
EHR + SNP (deep models) Prediction (CN, MCI, AD) | Regularization coefficients (0.03, 0.03) Dropouts (0.6, 0.6, 0.6) Layer sizes (200, 100, 50) Random Forest Trees = 31 | Biomarker (Asymmetric dimethylarginine) Neuropsychological Battery (AVERR total intrusions) Neuropsychological Battery (Auditory Verbal Learning Test Trial1) Memory Score Voxel based morphometry Amygdala left | Accuracy: 0.78 ± 0 Precision: 0.79 ± 0.07 Recall: 0.79 ± 0.09 F1 Scores: 0.79 ± 0.07 | Accuracy: 0.78 Precision: 0.78 Recall: 0.79 F1 Scores: 0.78 |
EHR + Imaging (deep models) Prediction (CN, MCI, AD) | Regularization coefficients (0.03, 0.03) Dropouts (0.6, 0.6, 0.6) Layer sizes (200, 100, 50) Random Forest Trees = 31; | Biomarker (Asymmetric dimethylarginine) Neuropsychological Battery (AVERR total intrusions) Cortical Thickness Average of Right Pericalcarine Memory Score Voxel based morphometry Amygdala left | Accuracy: 0.79 ± 0 Precision: 0.79 ± 0.08 Recall: 0.79 ± 0.08 F1 Scores: 0.79 ± 0.07 | Accuracy: 0.77 Precision: 0.76 Recall: 0.77 F1 Scores: 0.77 |
SNP + Imaging (shallow models) Prediction (CN, MCI/AD) | Random Forest Trees = 20 | Mean GLCM 3 right superior temporal Sum GLCM 5 left amygdala Median GLCM 2 right hippocampus Gene10 location 108777098 Entropy intensity left hippocampus | Accuracy: 0.75 ± 0.11 Precision: 0.72 ± 0.16 Recall: 0.65 ± 0.09 F1 Scores: 0.65 ± 0.12 | Accuracy: 0.63 Precision: 0.62 Recall: 0.57 F1 Scores: 0.56 |
