Table 1.
Summary of work and predictions relating to the detection of SZ using data from structural MRI scans via various artificial intelligence techniques and machine learning algorithms.
| Study | Year | Subjects | Prediction | AI/ML Technique | |
|---|---|---|---|---|---|
| Patients | Control | ||||
| Leonard et al. [26] | 1999 | 37♂ | 33♂ | 77% | Linear Discriminant Function Analysis (DFA) |
| Csernansky et al. [31] | 2002 | 52 | 65 | 75% (sensitivity) 76.9% (specificity) |
Logistic Regression Model |
| Nakamura et al. [33] | 2004 | 30♂, 27♀ | 25♂, 22♀ | 80%♂, 81.6%♀ | DFA |
| Yushkevich et al. [34] | 2005 | 46 | 46 | 72% (sensitivity) 70% (specificity) |
Support Vector Machine (SVM) |
| Davatzikos et al. [32] | 2005 | 69 | 79 (matched) | 81.1% (mixed) 85%♂, 82%♀ |
High-dimensional nonlinear Pattern Classifier |
| Fan et al. [35] | 2006 | 23♀, 46♂ | 38♀, 41♂ | 91.8%♀, 90.8%♂ | Nonlinear SVM, leave-one-out cross-validation |
| Yoon et al. [36] | 2007 | 21♀, 32♂ | 52 (matched) | at least 88.8% | SVM, PCA |
| Kawasaki et al. [37] | 2007 | 30♂, 16♂ | 30♂, 16♂ | 90%, 80%, 75% (Jackknife) |
Multivariate Linear DFA, Jackknife approach |
| Castellani et al. [38] | 2009 | 54 | 54 | up to 75% and 85% (sex stratified) | Scale Invariance Feature Transform (SIFT), SVM |
| Pohl and Sabuncu [39] | 2009 | 16 | 17 (age-matched) | up to 90% | Linear SVM, Leave-one-out cross-validataion |
| Sun et al. [40] | 2009 | 36 | 36 (sex- and age-matched) | 86.1% | Pattern Classification Analysis with Sparese Multi-nomial Logistic Regression Classifier, Leave-on-out cross-validation |
| Koutsouleris et al. [27] | 2009 | A1: 20 (ARMS-E), 25 (ARMS-L) A2: 15 (ARMS-T), 18 (ARMS-NT) |
A1: 25 (matched) A2: 17 (matched) Cross-validation: 45 |
at least 86% (sensitivity) at least 93% (specificity) |
SVM, Multivariate Pattern Analysis (MVPA) |
| Takayanagi et al. [41] | 2010 | 17♂, 17♀ | 24♂, 24♀ | 75.6%, 82.9% | Linear DFA |
| Castellani et al. [42] | 2010 | 64 | 60 | up to 86.13% | SVM |
| Koutsouleris et al. [43] | 2010 | 25 | 28 | 83% | SVM with Partial-least-squares Pattern Analysis |
| Kasparek et al. [44] | 2011 | 39 | 39 | 66.7% (sensitivity) 76.9% (specificity) |
Maximum-uncertainty Linear Discriminant Analysis (MLDA) |
| Karageorgiou et al. [45] | 2011 | 28 | 47 | 67.9% (sensitivity) 72.3% (specificity) using PCA-LDA (sMRI only) |
LDA, Principal Component Analysis (PCA) |
| Castellani et al. [46] | 2011 | 30 | 30 | up to 83.33% | SVM, Leave-one-out cross-validation |
| Ulaş et al. [47] | 2011 | 64 | 60 | 71.93% (SVM) | 1-Nearest Neighbour, Linear SVM |
| Koutsouleris et al. [48] | 2012 | 16/21 | 22 | 92.3% 66.9% 84.2% |
SVM |
| Castellani et al. [49] | 2012 | 54 | 54 (matched) | at least 66.38% | SIFT and nonlinear SVM |
| Nieuwenhuis et al. [50] | 2012 | 128, 155 | 111, 122 | 71.4%, 70.4% | SVM, Leave-one-out cross-validation |
| Ulaş et al. [28] | 2012 | 50 | 50 | 84% (MKL) 77% (SVM) |
SVM, MKL |
| Ulaş et al. [29] | 2012 | 21♂, 21♀ | 19♂, 21♀ | 90.24% (CLMKL) 71.95% (SVM) |
SVM, Clustered Localized MKL (CLMKL) |
| Ota et al. [51] | 2012 | 38♀, 23♀ | 105♀, 23♀ | 74% (sensitivity) 70% (specificity) |
DFA |
| Bansal et al. [52] | 2012 | 65 | 40 | 93.1% (sensitivity) 94.5% (specificity) |
Hierarchical clustering, Split-half and Leave-one-out cross-validation |
| Greenstein et al. [53] | 2012 | 98 | 99 | 73.3% | Random Forest |
| Borgwardt et al. [54] | 2013 | 16/23 | 22 | 86.7% 80.7% 80.0% |
SVM, Nested cross-validation |
| Iwabuchi et al. [55] | 2013 | 19 | 20 | up to 77% | SVM |
| Zanetti et al. [56] | 2013 | 62 | 62 (matched) | 73.4% | SVM |
| Gould et al. [57] | 2014 | 126/74 | 134 | 71% | SVM |
| Perina et al. [58] | 2014 | 21♂, 21♀ | 19♂, 21♀ | 83% (sensitivity) | SVM |
| Schnack et al. [59] | 2014 | 46/47 | 43 | 90% | SVM |
| Cabral et al. [60] | 2016 | 71 | 74 | 69.7% | SVM, MVPA |
| Lu et al. [61] | 2016 | 41 | 42 (sex- and age-matched) | 91.9% (sensitivity) 84.4% (specificity) |
SVM, Recursive Feature Elimination (RFE) |
| Yang et al. [30] | 2016 | 40 | 46 | 77.91% | MLDA, SVM |
| Squarcina et al. [62] | 2017 | 127 | 127 | 80% | SVM |
| Rozycki et al. [63] | 2018 | 440 | 501 | 76% | Linear SVM |
| de Moura et al. [64] | 2018 | 143, 32 | 82 | 77.6% (sensitivity) 68.3% (specificity) |
MLDA |
| Liang et al. [65] | 2019 | 98, 54 | 106, 48 | 75.05%, 76.54% | Gradient Boosting Decision Tree |
| Deng et al. [66] | 2019 | 65 | 60 | 76.9% (sensitivity) 75.0% (specificity) |
Random Forest |