Table 3.
Overview of ADHD studies included in this literature review.
| Authors, year, ref | Dataset | Sample size | # Sites | Input Modality | Feature selection (y/n) | Feature engineering | # Features | Task | Validation | Best DL model | DL ACC | Best ML model | ML input | ML ACC |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Kuang et al. 2014, (Kuang et al., 2014) | ADHD-200-NYU | HC = 107, ADHD-C = 99, ADHD-I = 44, ADHD-H = 13 | 1 | rs-fMRI | yes (expert) | ROI (PFC) max freq | x | HC vs. ADHD-C vs ADHD-I vs ADHD-H | Train/test | DBN | 37.4 | x | x | x |
| Kuang and He, 2014, (Kuang and He, 2014) | ADHD-200 | HC = 160, ADHD-C = 125, ADHD-I = 50, ADHD-H = 14 | 3 | rs-fMRI | no | WB freq PCA | 257*9177 | HC vs. ADHD-C vs ADHD-I vs ADHD-H | Train/test | DBN | 44.6 | x | x | x |
| Hao et al., 2015, (Hao et al., 2015) | ADHD-200_NYU | HC = 110, ADHD-C = 95, ADHD-I = 2, ADHD-H = 50 | 1 | rs-fMRI | no | selected ROI network of 14 ROIS | x | HC vs. ADHD-C vs ADHD-I vs ADHD-H | 100 cv | DBaN | 64.7 | x | x | x |
| Sen et al., 2018*, (Sen et al., 2018) | ADHD-200 | ADHD = 356, HC = 373 | 8 | s-MRI & rs-fMRI | no | Unsupervised features (structural + spatio-temporal) | 45 IC for fMRI + sMRI | ADHD vs TPC | Train/test | Multimodal feature learning + linear SVM | 67.3 | x | x | x |
| Wang & Kamata, 2019, (Wang and Kamata, 2019) | ADHD-200 | ADHD = 362, HC = 585 | 7 | s-MRI | no | 3D fractal dimension complexity map (FDCM) | 96*120*100 | ADHD vs TPC | Train/test | 3D CNN | 69.0 | x | x | x |
| Zou et al. 2017, (Zou et al., 2017) | ADHD-200 | ADHD = 197, HC = 362 | 8 | rs-fMRI | no | ReHo, fALFF, VMHC | 3 * 47 * 60 * 46 + 3 * 90*117 *100 | ADHD vs TPC | 10 cv and leave-site out | 3D CNN | 69.2 | x | x | x |
| Riaz et al., 2018, (Riaz, et al., 2018) | ADHD-200 | HC = 95, ADHD-C = 127* | 1 | rs-fMRI | no | 90 AAL timeseries | 900*T | HC vs ADHD | Train/test | CNN | 73.1 | SVM | FC matrix with feature selection of elastic net | 56.1 |
| Wang et al., 2019**, (Wang et al., 2019) | ADHD-200 | ADHD = 146, HC = 441 | 8 | s-MRI | no | full 3D image | 121*145*121 | ADHD vs TPC | 5 cv | 3D CNN | 76.6 | x | x | x |
| Desphande et al., 2015, (Deshpande et al., 2015) | ADHD-200 | HC = 744, ADHD-C = 260, ADHD-I = 173 | 7 | rs-fMRI | yes, (PCA) | 200 PCA connectivity features | 20 | HC vs ADHD-C | LOOCV | Fc cascade NN with 2 training stages | ~90 | SVM | significant features of PCA + conn weights | ~80 |
* General model for ADHD and ASD, ** General model for ADHD and SZ
aAUC ROC, b F score, c Balanced accuracy
! not clear if feature selection is done only on training set, !! Feature selection done before train/test split
ADHD = Attention Deficit hyperactivity disorder, -I, Inattentive, -H hyperactive, -C combined, HC = healthy control, rs = resting state, fMRI = functional Magnetic Resonance Imaging, s-MRI = structural Magnetic Resonance Imaging, AAL = automatic anatomic labelling, ROI = Region of interest, CNN = convolutional neural network, DBN = Deep belief network, DBaN = deep baysesian network, FC = functional connectivity, SVM = support vector machine, 10 cv = 10 fold cross validation, LOOCV = leave one out cross validation, PCA = principle component analysis, ReHO = regional homogeneity, VHMC = voxel-mirrored homotopic connectivity, fALFF = Fractional amplitude of low-frequency fluctuations, NN = neural network.