Table 1.
Description of the studies.
| Year | Author [ref.] | Study design | Intervention | Purpose | Objective | Sample size | AI accuracy for best model |
|---|---|---|---|---|---|---|---|
| 2022 | Bendifallah et al.50 | Retrospective | Logistic Regression, Random Forest, Decision Tree, eXtreme Gradient Boosting, Voting Classifier (soft/hard) | Prediction | Predict likelihood of endometriosis based on 16 essential clinical and symptom-based features related to patient history, demographics, endometriosis phenotype and treatment | 1126 endometriosis patients, 608 controls |
SE = 93% SP = 92% |
| 2022 | Bendifallah et al.35 | Prospective | Logistic Regression, Random Forest eXtreme Gradient Boosting, AdaBoost | Diagnosis | Diagnosis of endometriosis using a blood-based mRNA diagnostic signature | 200 plasma samples (153 cases, 47 controls) |
SE = 96.8% SP = 100% |
| 2021 | Maicus et al.61 | Prospective | Resnet (2 + 1)D | Diagnosis | Classification of the state of the Pouch of Douglas using the sliding sign test on ultrasound | 749 transvaginal ultrasound videos (414 training set, 139 validation set, 196 test set) |
SE = 88.6% SP = 90% |
| 2021 | Guerriero et al.59 | Retrospective | K-Nearest Neighbor, Naïve Bayes, Neural Networks, SVM, Decision Tree, Random Forest, Logistic Regression | Prediction | Detection of endometriotic bowel involvement in rectosigmoid deep endometriosis | 333 patients |
SE = 72% SP = 73% |
| 2021 | Li et al.52 | Retrospective | Deep Machine Learning Algorithm (NNET) | Diagnosis | Diagnosis of endometriosis based on genes | 213 patients |
SE = 100% SP = 61.1% |
| 2020 | Matta et al.30 | Retrospective Case–Control | Logistic Regression, ANN, SVM, Adaptive Boosting, PLSDA | Research | Identify biomarkers of internal exposure in adipose tissue most associated with endometriosis | 99 women (44 controls, 55 cases) |
SE = NR SP = NR |
| 2020 | Akter et al.53 | Retrospective | New Ensemble Machine Learning Classifier (GenomeForest) | Diagnosis | Classifying endometriosis versus control patients using RNAse and enrichment-based DNA-methylation datasets | 38 single-end RNA-sequence samples, 80 MBD-sequence DNA-methylation samples |
Transcriptomics Data SE = 93.8% SP = 100% Methylomics Data SE = 92.9% SP = 88.6% |
| 2020 | Perrotta et al.54 | Prospective Observational Cross-Sectional Pilot | Random Forest-Based Machine Learning Classification Analysis | Diagnosis | Diagnosis of endometriosis using gut and/or vaginal microbiome profiles | 59 women (24 controls, 35 endometriosis patients) |
SE = NR SP = NR |
| 2020 | Guo et al.58 | Retrospective Cohort | Logistic Regression | Prediction | Predict any-stage and stage 3/4 endometriosis before surgery in infertile women | 1016 patients (443 without endometriosis, 377 patients with stage 1/2 endometriosis, 196 patients with stage 3/4 endometriosis) |
SE = NR SP = NR |
| 2021 | Vesale et al.45 | Retrospective | Logistic Regression | Prediction | Predict likelihood of voiding dysfunction after surgery for deep endometriosis | 789 patients |
SE = NR SP = NR |
| 2019 | Benoit et al.46 | Retrospective | Logistic Regression | Prediction | Predict likelihood of a live birth after surgery followed by ART for patients with endometriosis-related infertility | 297 women |
SE = NR SP = NR |
| 2019 | Lee et al.29 | Retrospective | Recommendation System | Research | Identify diseases associated with endometriosis | 1,730,562 controls, 11,273 cases |
SE = NR SP = NR |
| 2019 | Braga et al.36 | Prospective Case–Control | PLSDA | Diagnosis | Develop an adjuvant tool for diagnosis of grades 3 and 4 endometriosis in infertile patients | 50 endometriosis serum samples, 50 control samples |
SE = NR SP = NR |
| 2019 | Chattot et al.57 | Prospective Observational | Logistic Regression | Prediction | Predict rectosigmoid involvement in endometriosis using preoperative score | 119 women undergoing surgery for endometriosis |
SE = NR SP = NR |
| 2019 | Knific et al.31 | Retrospective | Decision Tree, Linear Model, K-Nearest Neighbor, Random Forest | Diagnosis | Diagnosis of endometriosis based on plasma levels of proteins and patients’ clinical data | 210 patients |
SE = 40% SP = 65% |
| 2019 | Parlatan et al.37 | Retrospective | K-Nearest Neighbor, SVM, PCA | Diagnosis | Diagnosis of endometriosis using non-invasive Raman spectroscopy-based classification model | 94 serum samples (49 endometriosis, 45 controls) |
SE = 89.7% SP = 80.5% |
| 2019 | Akter et al.55 | Retrospective | Decision Tree, PLSDA, SVM, Random Forest | Diagnosis | Classify endometriosis versus control biopsy samples using transcriptomics or methylomics data | 38 samples in transcriptomics dataset, 77 samples in methylomics dataset |
Transcriptomics Data SE = 81.3% SP = 95.5% Methylomics Data SE = 76.2% SP = 80% |
| 2018 | Bouaziz et al.28 | Retrospective | NLP | Research | Using NLP to extract data by text mining of the endometriosis-related genes in the PubMed database | 724 genes retrieved |
SE = NR SP = NR |
| 2017 | Dominguez et al.33 | Prospective Case–Control | SVM | Diagnosis | Diagnosis of endometriosis using lipidomic profiling of endometrial fluid in patients with ovarian endometriosis | 12 endometriosis, 23 controls |
SE = 58.3% SP = 100% |
| 2016 | Ghazi et al.38 | Prospective Cohort | PLSDA, Multi-Layer Feed Forward ANN, QDA | Prediction | Determine classifier metabolites for early prediction risk of disease | 31 infertile women with endometriosis, 15 controls |
SE = NR SP = NR |
| 2015 | Reid et al.60 | Prospective Observational | Logistic Regression | Prediction | Use mathematical ultrasound models to determine whether a combination of transvaginal sonography markers could improve prediction of Pouch of Douglas obliteration | 189 women with suspected endometriosis |
Model 1 SE = 88% SP = 97% Model 2 SE = 88% SP = 99% |
| 2014 | Lafay Pillet et al.47 | Prospective | Logistic Regression | Diagnosis | Diagnose DE before surgery for patients operated on for endometriomas | 164 patients with DIE, 162 with no DIE |
SE = 51% SP = 94% |
| 2014 | Tamaresis et al.56 | Retrospective | Margin Tree Classification | Diagnosis | Detect and stage pelvic endometriosis using genomic data from endometrium | 148 endometrial samples |
SE = NR SP = NR |
| 2014 | Wang et al.39 | Prospective Case–Control | Genetic Algorithm, Decision Tree Algorithm, Quick Classifier Algorithm | Diagnosis | Diagnosis of endometriosis and stage using peptide profiling | 122 patients |
SE = 90.9% SP = 92.9% |
| 2013 | Wang et al.51 | Retrospective | Decision Tree | Prediction | Predict medical care decision rules for patients with recurrent pelvic cyst after surgical interventions | 178 case records |
SE = NR SP = NR |
| 2012 | Ballester et al.48 | Prospective Longitudinal Study | Logistic Regression | Prediction | Prediction of clinical pregnancy rate in patients with endometriosis | 142 infertile patients with DIE |
SE = 66.7% SP = 95.7% |
| 2012 | Fassbender et al.40 | Retrospective | LSSVM | Diagnosis | Diagnosis of endometriosis undetectable by ultrasonography | 254 plasma samples (89 controls, 165 endometriosis patients) |
SE = 88% SP = 84% |
| 2012 | Fassbender et al.41 | Retrospective | LSSVM | Diagnosis | Diagnosis of endometriosis through mRNA expression profiles in luteal phase endometrium biopsies | 49 endometrial biopsies |
SE = 91% SP = 80% |
| 2012 | Vodolazkaia et al.34 | Retrospective Cohort | Logistic Regression, LSSVM | Diagnosis | Diagnosis of endometriosis in symptomatic patients without U/S evidence of endometriosis | 121 controls, 232 endometriosis patients |
SE = 81% SP = 81% |
| 2012 | Dutta et al.42 | Prospective | PLSDA | Prediction | Identification of predictive biomarkers in serum for early diagnosis of endometriosis in a minimally invasive manner | 22 endometriosis, 23 controls |
SE = 81.8% SP = 91.3% |
| 2012 | Nnoaham et al.27 | Prospective Observational | Logistic Regression | Prediction | Predict any-stage endometriosis and stage 3 and 4 disease with a symptom-based model | 1396 symptomatic women |
SE = 82.6% SP = 75.8% |
| 2010 | Wang et al.26 | Retrospective | ANN | Prediction | Screening for biomarkers of eutopic endometrium in endometriosis patients | 26 patients |
SE = 91.7% SP = 90.9% |
| 2009 | Wolfler et al.43 | Prospective Exploratory Cohort | Genetic Algorithm | Prediction | Predict endometriosis before laparoscopy using patterns of serum proteins in symptomatic patients | 91 symptomatic patients |
SE = 81.3% SP = 60.3% |
| 2009 | Stegmann et al.62 | Prospective Cohort | Logistic Regression | Prediction | Prediction of lesions that have high probability of containing histologically-confirmed endometriosis | 114 women with complete data on 487 lesions |
SE = 88.4% SP = 24.6% |
| 2008 | Wang et al.44 | Retrospective | ANN | Diagnosis | Diagnostic model to correctly detect endometriosis and no endometriosis in serum samples using potential biomarkers of endometriosis | 66 serum samples |
SE = 91.7% SP = 90% |
| 2005 | Chapron et al.49 | Prospective | Logistic Regression | Prediction | Predict presence of posterior deep endometriosis among women with chronic pelvic pain symptoms | 134 women scheduled for laparoscopy for chronic pelvic pain symptoms |
SE = 68.6% SP = 77.1% |
NR not reported, PLSDA partial least squares discriminant analysis, QDA quadratic discriminant analysis, SVMs support vector machines, ANNs artificial neural networks, LSSVMs least squares support vector machines, PCA principal component analysis, NLP natural language processing, DE deep endometriosis, U/S ultrasound, miRNAs microRNAs, ART assisted reproductive technology, RNA ribonucleic acid, DNA deoxyribonucleic acid, MBD methyl binding domain, SE sensitivity, SP specificity.