Machine learning algorithms as new screening approach for patients with endometriosis

Abstract

Endometriosis—a systemic and chronic condition occurring in women of childbearing age—is a highly enigmatic disease with unresolved questions. While multiple biomarkers, genomic analysis, questionnaires, and imaging techniques have been advocated as screening and triage tests for endometriosis to replace diagnostic laparoscopy, none have been implemented routinely in clinical practice. We investigated the use of machine learning algorithms (MLA) in the diagnosis and screening of endometriosis based on 16 key clinical and patient-based symptom features. The sensitivity, specificity, F1-score and AUCs of the MLA to diagnose endometriosis in the training and validation sets varied from 0.82 to 1, 0–0.8, 0–0.88, 0.5–0.89, and from 0.91 to 0.95, 0.66–0.92, 0.77–0.92, respectively. Our data suggest that MLA could be a promising screening test for general practitioners, gynecologists, and other front-line health care providers. Introducing MLA in this setting represents a paradigm change in clinical practice as it could replace diagnostic laparoscopy. Furthermore, this patient-based screening tool empowers patients with endometriosis to self-identify potential symptoms and initiate dialogue with physicians about diagnosis and treatment, and hence contribute to shared decision making.

Revisiting the Risk Factors for Endometriosis: A Machine Learning Approach

“…With the increasing use of medical imaging, videos, and pathological samples, machine learning and deep learning approaches are playing a growing role in diagnosis [ 61 ]. A machine learning model for endometriosis based on a screening questionnaire was shown to produce an AUC of 0.5–0.9 in the training and validation sets based on the combination of 16 common criteria such as age, pain, and family history [ 62 ]. We demonstrated that the reanalysis of large cohorts of diagnosed women with endometriosis from the general population of UKB provided attributes and measurements not traditionally associated with the disease, and which were not informative under standard univariate statistical tests.…”

Unified Predictive Model for Endometriosis: Merging Clinical, Self-reporting and Genetic Information

“…With the increasing use of medical imaging, videos, and pathological samples, machine learning and deep learning approaches are playing a growing role in diagnosis [55]. A machine learning model for endometriosis based on a screening questionnaire was shown to produce an AUC of 0.5-0.9 in the training and validation sets based on the combination of 16 common criteria such as age, pain, and family history [56]. We show prediction of endometriosis in the general population of UKBB can use attributes and measurements not traditionally associated with the disease, and which were not informative under standard univariate statistical tests.…”

Clinical use of artificial intelligence in endometriosis: a scoping review

“…Diagnostic or predictive models for endometriosis using clinical variables and symptoms Six studies [45][46][47][48][49][50] grouped in this category strongly preferred using logistic regression; two studies 50,51 used decision tree methods to build a model and one study 50 also used random forest, eXtreme gradient boosting and voting classifier (soft/hard) ML algorithms as shown in Table 4. Interestingly many studies in this category examined predictive and diagnostic model capabilities in patients with some form of deep endometriosis (n = 5).…”

“…Interestingly many studies in this category examined predictive and diagnostic model capabilities in patients with some form of deep endometriosis (n = 5). The pooled SE for the models with highest accuracy in each study was 81.7% while the pooled SP was 91.6% [47][48][49][50] . Specific inputs into each model varied as seen in previous categories with Bendifallah et al 50 using the largest number of clinical features for their models.…”

“…The pooled SE for the models with highest accuracy in each study was 81.7% while the pooled SP was 91.6% [47][48][49][50] . Specific inputs into each model varied as seen in previous categories with Bendifallah et al 50 using the largest number of clinical features for their models. However, there were some commonalities in the types of inputs that were used in each model.…”