Table 4.
Application of machine learning based on imaging biomarker genomics in AD diagnosis and prognosis
| Method | Year | Modality | Model | Dataset | CV | Neural location | Results |
|---|---|---|---|---|---|---|---|
| Machine learning | 2010 [156] | sMRI, FDG PET, CSF, APOE genotype, age, sex, body mass index | SVM |
HC: 213 AD: 158 MCI: 264 |
LOOCV |
Hippocampal, ventricular, temporal lobe |
A maximum up to 90% accuracy for AD |
| 2013 [155] | sMRI, FDG PET, CSF, APOE genotype | MRF |
HC: 35 AD: 37 MCI: 75 |
Fourfold CV | Whole brain | An accuracy of 89% for AD | |
| 2014 [164] |
sMRI, FDG PET, CSF, SNP |
SVM |
HC: 47 AD: 49 MCI: 93 |
Tenfold CV |
Whole brain | An accuracy of 71% among HC, MCI and AD | |
| 2016 [157] | APOE genotype, neuropsychological assessment, sMRI, FDG PET | NB |
HC: 112 AD: 144 sMCI: 265 pMCI: 177 |
independent test set | Whole brain | An accuracy of 87% in identifying pMCI from sMCI | |
| 2017 [159] | sMRI, SNP | HYDRA |
HC: 139 AD: 103 |
– |
Hippocampus, entorhinal cortex frontal lobe |
The highest AUC value of 0.942 for AD | |
| 2017 [165] | sMRI, SNP | SVM |
HC: 204 AD: 171 MCI: 362 |
Tenfold CV |
Whole brain | An accuracy of 80.8% identifying pMCI from sMCI | |
| 2019 [158] | fMRI, SNP | MRF |
HC: 35 AD: 37 |
– | Olfactory cortex, insula, posterior cingulate gyrus and lingual gyrus | An accuracy of 87% AD prediction | |
| 2019 [154] | SNP |
LASSO, KNN, SVM |
HC: 371 AD: 267 |
CV | – | The highest reached 0.72 of the AUC | |
| 2019 [166] | APOE, PET, PGS | LR |
HC: 224 AD: 174 MCI: 344 |
– | Whole brain | An AUC value of 0.69 using PGS and APOE to predict amyloid state | |
| 2020 [167] | sMRI, FDG PET, AV45 PET, DTI, resting-state fMRI, APOE genotype | MKL |
HC: 35 AD: 33 sMCI: 30 pMCI: 31 |
LOOCV | Whole brain | An accuracy of 96.9% in identifying pMCI from sMCI | |
| Deep learning | 2017 [162] |
SNP, sMRI FDG PET |
DFFF |
HC: 226 AD: 190 MCI: 389 |
Twentyfold CV | Whole brain | An accuracy of 0.65 among HC, MCI and AD |
| 2018 [68] | sMRI, SNP | NN |
HC: 225 AD: 138 MCI: 358 |
Fivefold CV | 16 ROIs (hippocampus, entorhinal cortex, parahippocampal gyrus, amygdala, precuneus, etc.) | An AUC value of 0.992 using combined features | |
| 2019 [161] | sMRI, demographic, neuropsychological assessment, APOE genotype data | CNN |
HC: 184 AD: 192 sMCI: 228 pMCI: 181 |
Tenfold CV | Whole brain | An AUC value of 0.925 for pMCI prediction | |
| 2019 [160] | DTI, SNP | DCNN |
HC: 100 AD: 51 |
Fivefold CV | Temporal lobes (including the hippocampus) and the ventricular system | The highest AUC value of 0.858 | |
| 2021 [61] | MRI, SNP, electronic health records | CNN | ADNI | independent test set | Whole brain | A maximum up to 87% accuracy |
CNN convolutional neural network, CV cross validation, DCNN deep CNN, DFFF deep feature learning and fusion framework, HYDRA heterogeneity through discriminative analysis, LOOCV leave-one-out CV, MKL multiple kernel learning, MRF multimodal random forest, NN neural network, pMCI progressive MCI, sMCI stable MCI