Table 4.
AI performance and integration level for each AI application included in this review. AUC, area under the ROC curve; LR, likelihood ratio; MAE, mean absolute error; RMSE, root mean square error; R2, correlation coefficient
| Reference | Aim of the AI application | Type of AI algorithm | AI model performance | aTRL |
|---|---|---|---|---|
| Karasik et al. 1999 [32] | Age estimation for clinical forensic medicine purposes | Multilinear regression | R2 between estimated and ground truth values ranges between 0.818 and 0.901 | 2 |
| Karasik et al. 2000 [33] | Age estimation for clinical forensic medicine purposes | Logistic regression | R2 between estimated and ground truth values ranges between 0.671 and 0.901. Standard error estimate ranges between 4.22 and 6.64 years | 2 |
| Bocaz-Beneventi et al. 2002 [20] | Estimation of postmortem interval | Artificial neural network | Average residual of the difference between the estimated and experimental values on the validation set is 3.04 h | 2 |
| Constantinou et al. 2015 [46] | Assessment of the risk of violent reoffending | Bayesian network | AUC is 0.78 | 2 |
| Simmons et al. 2016 [13] | Postmortem identification | Decision tree | Accuracy is 1 | 2 |
| Stern et al. 2016 [34] | Age estimation for clinical forensic medicine purposes | Deep convolutional neural network | Best MAE is 0.36 ± 0.3 years | 2 |
| Yilmaz et al. 2017 [23] | Determination of the causes of death | Artificial neural network, logistic regression and radial-basis function network | Specificity is 0.833, sensitivity is 1, F score is 0.9091, accuracy is 0.9 | 2 |
| Ebert et al. 2017 [24] | Determination of the causes of death | Artificial neural network (architecture is not described) |
Detection task: average precision, sensitivity and F score are respectively 0.85 ± 0.11, 0.77 ± 0.26, 0.77 ± 0.16 Segmentation task: average precision, sensitivity, and F score are respectively 0.79 ± 0.05, 0.78 ± 0.05, 0.78 ± 0.0003 |
2 |
| Spampinato et al. 2017 [35] | Age estimation for clinical forensic medicine purposes | Convolutional neural network with regression network | MAE is 0.79 years | 2 |
| Stern et al. 2017 [36] | Age estimation for clinical forensic medicine purposes | Random forest and convolutional neural network (architecture is not described) |
Age estimation: MAE is 1.14 ± 0.96 years Majority age distinction: accuracy is 0.913, sensitivity is 0.886, and specificity is 0.932 |
2 |
| Zhang et al. 2018 [37] | Age estimation for clinical forensic medicine purposes | Linear regression, support vector machine, decision tree, and gradient boosting | MAE is 5.31 years for males and 6.72 years for females | 2 |
| Canturk et al. 2018 [21] | Estimation of postmortem interval | Support vector machine and k-nearest neighbors | Best accuracy is 0.89 | 2 |
| Heimer et al. 2018 [25] | Determination of the causes of death | Artificial neural network (architecture is not described) | AUC is 0.965, sensitivity is 0.914, and specificity is 0.875 | 2 |
| Koterova et al. 2018 [14] | Postmortem identification | Artificial neural network, decision tree, M5 tree, k-nearest neighbors, multilinear regression model, and collapsed regression model | MAE is 9.7 years and RMSE is 13.3 years | 2 |
| Matoba et al. 2018 [26] | Determination of the causes of death | Multivariate linear regression | R2 between estimated and real lung weight is 0.89 | 2 |
| Stern et al. 2019 [38] | Age estimation for clinical forensic medicine purposes | Convolutional neural network |
Biological age estimation: best MAE is 0.2 ± 0.42 years Chronological age estimation: best MAE is 0.82 ± 0.65 years Distinction of majority age: AUC is 0.9568 |
2 |
| Andersson et al. 2019 [22] | Estimation of postmortem interval | Bayesian network (architecture is not described) | LR < 1 | 2 |
| Avuclu et al. 2019 [15] | Age estimation and determination of gender for clinical forensic medicine purposes and postmortem identification | Multilayer perceptron |
Age estimation: difference between predicted and true age ranges from 0 to 6 years Gender determination: success rate between 1.5 and 100% depending on the method used to preprocess teeth images |
2 |
| De Back et al. 2019 [39] | Age estimation for clinical forensic medicine purposes | Bayesian convolutional neural network | Overall MAE is 21 months | 2 |
| Li et al. 2019 [40] | Age estimation for clinical forensic medicine purposes | Convolutional neural network | MAE is 0.89 years and RMSE is 1.21 years | 2 |
| Milosevic et al. 2019 [16] | Determination of gender for clinical forensic medicine purposes and postmortem identification | Convolutional neural network | Accuracy is 0.9687 ± 0.0096 | 2 |
| Turan et al., 2019 [17] | Determination of gender for clinical forensic medicine purposes and postmortem identification | Multilayer perceptron | Accuracy if 0.965, sensibility is 0.956, specificity is 0.973, and Matthews correlation coefficient is 0.929 | 2 |
| Abderrahmane et al. 2020 [41] | Age estimation for clinical forensic medicine purposes | Convolutional neural network combined with gated recurrent units | MAE is 1.9266 years | 2 |
| Garland et al. 2020 [27] | Determination of the causes of death | Convolutional neural network (architecture is not described) | Accuracy is 0.7 | 2 |
| Homma et al. 2020 [28] | Determination of the causes of death | Convolutional neural network | AUC is 0.879 | 2 |
| Peleg et al. 2020 [18] | Postmortem identification | Multivariate linear regression | Success rate ranges from 0.667 to 0.89 | 2 |
| Pena-Solorzano et al. 2020 [19] | Postmortem identification | Residual networks, hybrid convolutional auto-encoder and K-nearest neighbors |
Localization of femur: MAE, Jaccard similarity coefficient, and Dice score respectively range between 0 and 13.1 mm, 0.91 and 1, ranges between 0.93 and 1 Detection of implants: Accuracy, precision, recall, and F-score respectively range between 0.97 and 1, 0.91 and 0.99, 0.65 and 1, and 0.76 and 0.98 |
2 |
| Tirado et al. 2020 [47] | Bruise dating | Convolutional neural network | Sensitivity and precision are 0.97, and specificity is 0.995 | 2 |
| Vila-Blanco et al. 2020 [42] | Age estimation for clinical forensic medicine purposes | Convolutional neural network | R2 is 0.9, accuracy is 0.854, sensitivity is 0.878, specificity is 0.823, and AUC is 0.925 | 2 |
| Mauer et al. 2021 [43] | Age estimation for clinical forensic medicine purposes | Convolutional neural network + tree-based machine learning algorithm |
MAE is 0.71 ± 0.55 years for the coronal and 0.81 ± 0.62 years for the sagittal dataset Best accuracy, sensitivity, specificity, and AUC are respectively 0.875, 0.884, 0.886, and 0.943 for the sagittal dataset and 0.857, 0.864, 0.846, and 0.908 for the coronal dataset |
2 |
| Ozdemir et al. 2021 [44] | Age estimation for clinical forensic medicine purposes | Convolutional neural network | Kütahya Child Radiology Dataset: best MAE, RMSE, and R2 are 4.3, 5.76, and 0.99 respectively. Radiological Society of North America dataset: best MAE, RMSE, and R2 are 5.75, 7.42, and 0.96 respectively. The units of the performance metrics are not clear (years or months) | 2 |
| Oura et al. 2021 [29] | Determination of the causes of death | Multilayer perceptron | Testing accuracy and F1 range from 0.94 to 1, recall ranges from 0.89 to 1, precision from 0.92 to 1, and AUC from 0.99 to 1. Averaged test accuracy is 0.98 | 2 |
| Garland et al. 2021 [30] | Determination of the causes of death | Convolutional neural network | Accuracy and F1 scores are equal to 1 | 2 |
| Ibanez et al. 2022 [31] | Determination of the causes of death | Convolutional neural network | Recall, precision, and F1 score are respectively 0.93 ± 0.05, 0.89 ± 0.03, and 0.91 ± 0.04 | 2 |
| Li et al. 2022 [45] | Gender determination for clinical forensic medicine purposes | Convolutional neural network | Average accuracy is 0.946 in Chinese Han population and 0.829 in White population | 2 |