Table 2.
Studies with Predictions/Regression techniques.
| Author | Year | Data | Feature engineering | Learning Algorithm | Validation | Results |
|---|---|---|---|---|---|---|
| Abedin, J. [70] |
2019 | Questionnaire data /X-ray |
LASSO | Elastic Net (EN), Random Forests (RF) and a convolution neural network (CNN) |
70% training/30% testing | Root Mean Square Error (RMSE) for the CNN, EN, and RF models was 0.77, 0.97 and 0.94 respectively |
| Ashinsky, B. G. [68] |
2017 | MRI | – | Weighted neighbor distance using compound hierarchy of algorithms representing morphology WN(D-CHRM) | LOOCV | 75% acc |
| Donoghue, C. [65] |
2011 | MRI | Laplacian Eigenmap Embedding | Multiple linear regression | 270 knees as external validation group | Up to R2 = 0.75 |
| Du, Y. [69] |
2018 | MRI | PCA | ANN, SVM, Random forest, Naïve Bayes | 10-fold cross validation (10F-CV) | ANN with AUC = 0.761 for KL grade Random forest with area under the curve (AUC) = 0.785 for JSM |
| Du, Y. [67] |
2017 | MRI | PCA | ANN, SVM, Random forest, Naïve Bayes | 10F-CV | receiver operating characteristic (ROC) AUC of 0.761 (ANN) |
| Halilaj, E. [75] |
2018 | X-rays and pain scores | – | LASSO regression | 10F-CV for model selection and 10% for model evaluation | AUC of 0.86 for Radiographic progression |
| Lazzarini, N. [77] |
2017 | Clinical variables, food and pain questionnaires, biochemical markers (BM) and imaging-based information | Ranked Guided Iterative Feature Elimination, PCA | Random Forest | 10F-CV | AUC of 0.823 by using only 5 variables |
| Marques, J. [66] |
2013 | MRI | Texture Analysis for extraction and Partial least squares (PLS) regression for selection | Fisher linear discriminant analysis |
10F-CV for model selection. 10% for evaluation | ROC AUC of 0.92 |
| Nelson, A.E. [73] |
2019 | Demographic, MRI and biochemical variables |
Distance weighted discrimination (DWD), PCA | K- means, t-SNE | Validation on 597 participants- |
z = 10.1 (z-scores) |
| Pedoia, V. [71] |
2018 | MRI and biomechanics multidimensional data | Topological Data Analysis | Logistic Regression | – | AUC 83.8% |
| Tiulpin, Α. [74] |
2019 | X-ray, Clinical data | CNN | Logistic Regression (LR) and Gradient Boosting Machine (GBM) | OAI dataset for training and MOST dataset for testing, 5F-CV | AUC of 0.79 |
| Widera, P. [72] |
2019 | Clinical and X-ray image assessment metrics | Recursive feature elimination | Logistic regression, KNN, SVC (linear kernel), SVC (RBF kernel) and RF | Standard 10-fold stratified cross-validation protocol | F1 score 0.573–0.689 |
| Yoo, T. K. [76] |
2013 | Kinematic data | – | SVM | Leave-one-out cross-validation (LOOCV) | 97.4% acc |