Table 2.
Summary of work and predictions relating to the detection of SZ using data from functional MRI scans via various artificial intelligence techniques and machine learning algorithms.
| Study | Year | Subjects | Prediction | AI/ML Technique | |
|---|---|---|---|---|---|
| Patients | Control | ||||
| Jafri and Calhoun [68] | 2006 | 38 | 31 | 75.6% | Neural network |
| Calhoun et al. [69] | 2008 | 21 | 26 | 92% (sensitivity) 95% (specificity) |
MVPA |
| Anderson et al. [70] | 2010 | 14 | 6 | up to 90% | Multivariate Random Forest |
| Arribas et al. [71] | 2010 | 21 | 25 | 90% | Stochastic Gradient Learning based on minimization of Kullback-Leibler divergence |
| Shen et al. [72] | 2010 | 32 | 20 | 93.75% (sensitivity) 75% (specificity) |
Low-dimensional embedding and self-organized C-means clustering |
| Yang et al. [73] | 2010 | 20 | 20 | at least 82% (using fMRI data) | SVM |
| Castro et al. [74] | 2010 | 52 | 54 | 95% | Composite kernels, Linear and Gaussian SVM, Leave-two-out cross-validation |
| Costafreda et al. [75] | 2011 | 32 | 40 | 92% (seonsitivity) | SVM |
| Fan et al. [76] | 2011 | 31 | 31 | up to 85.5% | SVM, Linear kernel, Radial basis function kernel, Sigmoid kernel |
| Du et al. [77] | 2012 | 28 | 28 | 90% | Fisher’s linear discriminant analysis, Default mode network, Majority vote, Leave-one-out cross-validation |
| Liu et al. [78] | 2012 | 25 | 25 (siblings) 25 (HC) |
80.4% (SZ vs. HC) | Nonlinear SVM with polynomial kernel |
| Venkataraman et al. [79] | 2012 | 18 | 18 | 75% | Multivariate classification |
| Yoon et al. [80] | 2012 | 51 | 51 (age-matched) | 51.0% (sensitivity) 64.7% (specificity) |
Linear DFA, Leave-one-out cross-validation |
| Anderson and Cohen [81] | 2013 | 74 | 72 | 65% | SVM |
| Arbabshirani et al. [82] | 2013 | 28 | 28 | up to 96% (KNN) | Various (10 types) linear and nonlinear classifier |
| Fekete et al. [83] | 2013 | 8♂ | 10♂ | 100% | Complex network analysis, Block diagonal optimization. |
| Yu et al. [84] | 2013 | 24 | 25 (siblings) 22 (matched HC) |
62% | SVM, PCA, Leave-one-out cross-validation |
| Yu et al. [85] | 2013 | 32 (SZ) 19 (Depression) |
38 | 80.9% | SVM, Intrinsic DA, Leave-one-out cross-validation |
| Anticevic et al. [86] | 2014 | Sample: 90 Validation: 23 |
Sample: 90 (matched) Validation: 23 (matched) |
Sample: 75.5% (sensitivity), 72.2% (specificity) Validation: 67.9% (sensitivity), 77.8% (specificity) |
Linear SVM, Leave-one-out cross-validation |
| Brodersen et al. [87] | 2014 | 41 | 42 | 78%, 71% | Linear SVM, Variational Bayesian Gaussian mixture |
| Castro et al. [88] | 2014 | 31 | 21 | 90% (L-norm MKL), 85% (Lp-norm MKL) |
L-norm and Lp-norm MKL |
| Guo et al. [89] | 2014 | 69 | 62 | 68% | SVM |
| Watanabe et al. [90] | 2014 | 54 | 67 | at least 77.0% | Fused Lasso and GraphNet regularized SVM |
| Cheng et al. [91] | 2015 | 415 | 405 | 73.53–80.92% | SVM |
| Chyzhyk et al. [92] | 2015 | 26/14 | 28 | 97–100% | Linear SVM |
| Kaufmann et al. [93] | 2015 | 71 | 196 | 46.5% (sensitivity) 86.0% (specificity) |
Regularized LDA, Leave-one-out cross-validation |
| Pouyan and Shahamat [94] | 2015 | 10 | 10 | up to 100% (sensitivity and specificity) | ICA, PCA, Various, Leave-one-out cross-validation |
| Mikolas et al. [95] | 2016 | 63 | 63 (sex- and age-matched) | 74.6% (sensitivity) 71.4% (specificity) |
Linear SVM |
| Peters et al. [96] | 2016 | 18 | 18 | up to 91% | SVM, Leave-one-out cross-validation |
| Yang et al. [30] | 2016 | 40 | 40 | 77.91% | MLDA, SVM |
| Skaatun et al. [97] | 2017 | 182 | 348 | up to 80% | Multivariate regularized LDA |
| Chen et al. [98] | 2017 | 20 (SZ) 20 (depression) |
20 | 60% (sensitivity) 90% (specificity) |
Linear SVM, MVPA |
| Kaufmann et al. [99] | 2017 | 90 (SZ) 97 (bipolar) |
137 (HC) | 60% (sensitivity) 90% (specificity) |
5-class regularized LDA, k-fold cross-validation model |
| Guo et al. [100] | 2017 | 28 | 28 family-based control (FBC) 40 (HC) |
SVM: 96.43% (sensitivity) 89.29% (specificity, FBC) |
SVM, Receiver operating characteristic (ROC) curve |
| Iwabuchi and Palaniyappan [101] | 2017 | 71 | 62 | 80.32% | MKL |
| Yang et al. [102] | 2017 | 446 | 451 | 60–86% | Multi-task classification, 10-fold cross-validation |
| Bae et al. [103] | 2018 | 21 | 54 | 92.1% (SVM) | Various (5 types), 10-fold cross-validation |
| Li et al. [104] | 2019 | 60 | 71 | 76.34% (LDA) | KNN, Liner SVM, Radial basis SVM, LDA |
| Chatterjee et al. [105] | 2019 | 34 | 34 | 94% (SVM) 96% (1-NN) |
SVM, k-nearest neighbours |
| Kalmady et al. [106] | 2019 | 81 | 93 (sex- and age-matched) | 87% | L2-regularized Logistic regression |