Document Type : Original Article
Authors
1
Assisstant Professor of Department of Obstetrics and Gyenecology, Shahid Beheshti University of Medical Sciences, Shohadaye Tajrish Hospital, Tehran, Iran
2
Functional Neurosurgery research Center, Shohada Tajrish Neurosurgical Comprehensive Center of Excellence, Shahid Beheshti University of Medical Sciences, Tehran
3
School of medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran
4
Iran University of Science and Technology, Tehran, Iran
5
2. Functional Neurosurgery Research Center, Shohada Tajrish Comprehensive Neurosurgical Center of Excellence, Shahid Beheshti University of Medical Sciences, Tehran, Iran 5. USERN Office, Functional Neurosurgery Research Center, Shahid
10.22074/ijfs.2024.2009338.1519
Abstract
Background: Endometriosis is the presence of ectopic endometrial glands and is prevalent among women of childbearing age. It causes pelvic pain and infertility. The diagnosis of endometriosis is delayed in many patients. Definite diagnosis needs laparoscopic investigation although physical examination with ultrasonography can detect endometriosis with acceptable accuracy. Machine Learning (ML) helps with medical imaging. There was no sufficient ML study in Iranian female with endometriosis. In this study, we aimed to present the diagnostic accuracy of different ML algorithms for endometriosis.
Methods and materials: In this retrospective study we assessed the diagnostic accuracy of different ML algorithms in classifying suspicious cases of endometriosis by ultrasonographic signs. A data set of 505 patients (with 149 confirmed cases of endometriosis) was divided into training and test sets. We used stratified 5-fold cross-validation and area under the receiver operating characteristic curve (AUC) to train and validate the models, then Youden's J statistic to determine optimal thresholds. K-nearest Neighbors, Logistic Regression, Kernelized Support Vector Machines (SVM), Random Forest, Extremely Randomized Trees, Adaptive Boosting, and Gradient Tree Boosting models were developed.
Results: In the test set, 37 of 127 patients (29.1%) were diagnosed with endometriosis compared to 112 of 378 patients (29.6%) in the training set. The sensitivities and specificities ranged from 59.5% to 75.7% and 71.7% to 83.3%, respectively. The SVM, Random Forest, Extra-Trees, and Gradient Boosting models depicted the best performance with AUCs equal to 0.76.
Conclusion: In conclusion, our data support the use of ML models can be used can be beneficial for screening and diagnosis of endometriosis.
Keywords
Main Subjects
)