Table 3.
This table represents an overview of artificial intelligences, its respective performance, and data acquisition of each study.
| References | AI | Resolution | Data acquisition | MAE [SD], r2 |
|---|---|---|---|---|
| Ball et al. (2017) | Gaussian Process Regression | Not stated | 3 T, T1- and T2-weighted MRI & diffusion weighted scans | 1.54 y, r = 0.926 |
| Ball et al. (2021) | Regularized linear regression with elastic net penalty | Not stated | 3 T, T1-weighted MRI images | 1.81 y [0.06], 0.79 |
| Gaussian process regression | 1.75 y [0.07], 0.81 | |||
| Ensemble model | 1.92 y [0.10], 0.78 | |||
| Ball et al. (2019) | Gaussian Process Regression | Image resolution was 1 × 1 × 1.2 mm (Siemens) or 0.9375 × 0.9375 × 1.2 mm (GE) Diffusion data: resolution 2.5 × 2.5 mm (Siemens) or 1.875 × 1.875 (GE) and slice thickness = 2.5 mm |
3 T T1-weighted MRI & diffusion MRI | 2.18 y [0.18] |
| Brown et al. (2017) | Linear regression | 0.625 mm × 0.625 mm × 3 mm (Brown et al., 2015) | 1.5 T T1-weighted MRI & diffusion MRI scan | 6.284 w [4.230 SDAE] |
| Multi-layer perceptron | 7.223 w [5.080 SDAE] | |||
| SVR | 1.712 w [1.366 SDAE] | |||
| Bagging regression | 1.559 w [1.255 SDAE] | |||
| Random forests | 1.554 w [1.197 SDAE] | |||
| Brown et al. (2012) | Regularised multivariate nonlinear regression-like | T1: slice thickness = 1.2 mm T2: slice thickness = 2.5 mm |
3 T, T1- and T2-weighted MRI & diffusion weighted scans | 1.03 y, 0.92 |
| Cao et al. (2015) | LASSO/multivariate linear regression | GE scanners: slice thickness, 1.5 mm Siemens scanners: slice thickness, 1 mm |
1.5 T T1-weighted MRI | 1.69 y, r = 0.82 |
| Chen et al. (2022) | 3DCNN using T1 & T2 | Not stated | T1-, T2-weighted MRI | 7.7 w [1.7] |
| 3DCNN using T1 | 9.8 w [2.3] | |||
| 3DCNN using T2 | 9.1 w [1.9] | |||
| Chung et al. (2018) | Support vector regression with radial basis function kernel | 1.2-mm slices (256 × 192-mm) in-plane resolution (PING study) | 3 T T1-MRI weighted | 1.69 y, 0.84 |
| Erus et al. (2015) | SVR with a linear kernel | 0.9375 mm × 0.9375 mm × 1 mm | 3 T T1-MRI weighted & Diffusion Tensor Imaging MRI |
1.22 y |
| Franke et al. (2012) | RVR with a smoothing kernel | 1 × 1 × 1 mm3 1.5 mm or 1 mm (Siemens) slice thickness | 1.5 T, T1-MRI-weighted | 1.1 y, r = 0.93 |
| Galdi et al. (2020) | Linear regression model with elastic net regularisation | acquired voxel size = 1 mm isotropic T1 & T2 | 3 T T1 MRI-weighted & diffusion MRI | 0.7 w [0.56], r = 0.78 |
| Gschwandtner et al. (2020) | CNN | n.a. | EEG (8, 4 & 2 electrodes) | 8 EEG electrodes: 93.6% of estimations lying within ±2 w 67.9% within ±1 w deviation from PMA |
| He et al. (2020) | CNN | NIH-PD and MGHBCH | 1.5 T, T1-MRI | 0.96 y |
| Hosseinzadeh Kassani et al. (2020) | SVM classification with linear kernel | voxel size = 3 mm3 | Resting-state-functional MRI | 0.929 y [0.041] |
| Hu et al. (2020) | Hierarchical Rough-to-Fine model | T1: resolution with 1 × 1 × 1 mm3 T2: 1.25 × 1.25 × 1.95 mm3 |
3 T, T1-, T2-weighted MRI | 32.1 [1.2 days] |
| Hu et al. (2021) | Dimensional-attention-based 3D convolutional neural network | ABIDE I & II, ADHD200 | 3 T, T1-MRI | 1.01 y, MSE: 1.92y, 0.73 |
| Kardan et al. (2022) | SVR with a linear kernel | T1 & T2 slice thickness 0.8 mm | 3 T, T1- T2 MRI & rs-fMRI | 3.6 m, 0.51 |
| Kawahara et al. (2017) | CNN | Not stated | diffusion tensor imaging | 2.17 w [1.59] |
| Kelly et al. (2022) | Gaussian process regression | Victorian Infant Brain Study (VIBeS) | 3 T T1 MRI | 1.72 y [0.16], 0.80 |
| Khundrakpam et al. (2015) | Linear regression | 1 mm isotropic data GE scanner – 1.5 mm Fallback protocol 3 mm |
1.5 T, T1-MRI, T2-MRI, Proton Density | 1.68 y, r = 0.84 |
| Li et al. (2018) | CNN | Philadelphia Neurodevelopmental Cohort (PNC) | Rs-fMRI | 2.15y [1.54] R0.4 = 0.614 |
| Lund et al. (2022) | Linear regression | Healthy Brain Network (HBN) study sample & Philadelphia Neurodevelopmental Cohort (PNC) | 3 T T1 MRI & T2 rs-fMRI | 2.43 y [2.93], r = 0.6 |
| Morita et al. (2022b) | 3DCNN | n.a. | CT | 4.61 m [3.65], r = 0.89 |
| Morita et al. (2022a) | 3DCNN | n.a. | CT | RMSE: 6.45 m R = 0.89 Mean Prediction Error: 2.29 m [6.04] |
| Nielsen et al. (2019) | SVR | T1: 1 × 1 × 1 mm3 voxels | 3 T T1 MRI & rs-fMRI | R2 = 0.57 |
| O’Toole et al. (2016) | SVR with a linear kernel | n.a. | EEG (10 electrodes) | [7.85] d, r = 0.889 |
| Qu et al. (2020) | 3DCNN | ABIDE, ADHD200, HBN resampled to 1.5 × 1.5 × 1.5 mm3 | T1 MRI | 1.11, 0.78 |
| Saha et al. (2018) | 2DCNN | field of view 224 × 224 mm, matrix 128 × 128 in plane res: 1.75 × 1.75 mm, slice thickness not given |
Diffusion MRI | R = 0.6 |
| Shabanian et al. (2019) | 3DCNN | NIMH Data Archive (NDA) | 1.5 T, T1- T2-MRI & Proton density MRI | Class, Precision, F1-Score New-born, 1.00, 1.00; 1 Year, 0.95, 0.97; 3 Years, 1.00, 0.99 |
| Smyser et al. (2016) | SVM with a linear kernel | voxel size 1 × 1 × 1 mm3 | T2-MRI, rs-fMRI | Preterm vs. Term classification: 84% accuracy, 90% sensitivity and 78% specificity |
| Stevenson et al. (2017) | SVR | n.a. | EEG (9 electrodes) | R = 0.936 |
| Stevenson et al. (2020) | SVM | n.a. | EEG (9 electrodes) | Random error = 1.1 w Systematic error = −0.1 w |
| Vandenbosch et al. (2019) | Random Forest | n.a. | EEG (30 electrodes) | 1.22 y |
| RVM | 1.46 y | |||
| SVM | Not presented | |||
| Zhao T. et al. (2019) | SVM with a linear kernel | ADHD-200 T1: acquisition matrix: 256 × 256, FOV: 256 × 256 mm2; slice thickness – slice thickness 1.33 mm 1 × 1 × 1.33 mm Beijing cohort T1: in-plane resolution 1.0 × 1.0 mm, slice thickness 1.0 mm T2: in-plane resolution—0.7 × 0.7 mm, slice thickness—0.7 mm, |
3 T, T1- T2-MRI | r = 0.48 |
| RVM with a linear kernel | r = 0.48 | |||
| Sturmfels et al. (2018) | 3DCNN | Voxel size 0.94 × 0.94 × 1, FOV dimensions 196 × 256 × 160 | T1-MRI | 1.43 [0.03] |
| Hong et al. (2020) | 3D CNN | “newborns (≤1 month): voxel dimensions = 1.0 × 0.7 × 4.5 mm older children (>1 month): voxel dimensions = 1.4 × 1.0 × 5.0 mm” |
1.5 T T1-w MRI | 67.6d, 0.971 |
| Zhao Y. et al. (2019) | ridge regression | Not stated | 1.5 T or 3 T T1-weighted | 1.41 years, 0.71 1.42 years, 0.70 |
| Liang et al. (2019) | Penalized ridge regression | Not stated | T1w MRI | 6–30 years age range, 2.53 years, 0.85 |
| Support vector regression | Not presented | |||
| Gaussian processes regression | Not presented | |||
| Deep neural network | Not presented | |||
| Lewis et al. (2018) | Elastic net penalized linear regression model | resolution of 1 mm isotropic | 1.5 T T1-w MRI 3 T T1-w MRI |
504 d best model |
| Dean et al. (2015) | Voxel-wise probabilistic model | Not stated | Voxel-wise VFM maps | Males: 79.06 d Females: 90.02 d |
| Pardoe and Kuzniecky (2018) | Relevance vector machine regression | Not stated | T1-w MRI | 7.2 y |
| Gaussian processes regression | 8.4 y | |||
| Lavanga et al. (2018) | Linear mixed effect regression model | n.a. | EEG | 1.51 w, 0.8 |
| Liu et al. (2024) | Graph Convolutional Network (GCN) |
UCSF: enrolled until 2011: 1× 1 × 1 mm3 resolution enrolled between 2011 and 2017: 0.7 × 0.7 × 1 mm3 resolution |
UCSF: Enrolled until 2011: 1.5 T T1-w MRI enrolled between 2011 and 2017: 3 T T1-w MRI |
0.963 weeks, 0.94 |
|
dHCP: 0.5 × 0.5 × 0.5 mm3 resolution |
dHCP: 3 t T1-w MRI |
|||
| Tang et al. (2023) | 2D Convolutional Neural Network | Not stated | T1-w MRI | 1.15 |
| 3D Convolutional Neural Network | 1.80 | |||
| Liu et al. (2023) | radiomics first-order grayscale feature extraction method gray matter | Not stated | 1.5 T MRI | 104.41 |
| radiomics first-order grayscale feature extraction method white matter | 92.72 | |||
| FreeSurfer feature extraction method | 81.83 | |||
| Mendes et al. (2023) | Convolutional Neural Network | Not stated | T1-weighted MRI | 0.47 [0.01], 0.18 [0.04] |
| Zandvoort et al. (2024) | Support Vector Regression with linear kernel function | n.a. | EEG & EMG | 1.75 weeks (95% at [1.51, 2.03]) |
| Nielsen et al. (2023) | Support Vector Regression | 0.8-mm isotropic resolution | 3 T T2-weighted MRI & resting-state fMRI | R2: 0.51 0.59 |
| Hu et al. (2023) | Convolutional Neural Network on RAW data | 0–6 months: slice thickness = 4.0 mm, in-plane resolution = 0.7 × 0.7 mm2 6–36 months: slice thickness = 5.0 mm, in-plane resolution = 0.7 × 0.7 mm 2 |
3 T T1-weighted MRI | 67.66, 0.91 |
| Convolutional Neural Network on white matter | 72.17, 0.89 | |||
| Griffiths-King et al. (2023) | Gaussian Processes Regression | Not stated | T1w MRI | 1.48y, 0.37 |
| Bellantuono et al. (2021) | Deep neural network on the full dataset | Not stated | T1 weighted MRI | 2.19 [0.03y], 2.91 [0.03y] |
| Deep neural network on the subset of subjects within the 7–20 age range | 1.53 [0.02], 1.94 [0.02] | |||
| Deep neural network on an external dataset | 2.7 [0.2], 3.7 [0.2] |
y, year; m, month; w, week; MAE, Mean Absolute Error; rs-fMRI, resting-state functional MRI; RVM, Relevance vector machine; RVR, Relevance vector regression; RVM, Relevance vector machine; SVM, Support Vector Machine; SDAE, absolute error standard deviation; MSE, mean standard error.