Applicability of artificial intelligence in the analysis of clinical data and prediction of endometriosis during laparoscopy in women with chronic pelvic pain

• Grant number: 21/10074-8
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: January 01, 2022
• End date: December 31, 2022
• Field of knowledge: Health Sciences - Medicine
• Principal Investigator: Omero Benedicto Poli Netto
• Grantee: Mateus Carvalho Azevedo
• Host Institution: Faculdade de Medicina de Ribeirão Preto (FMRP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil

Abstract

Endometriosis is a common disease associated with negative repercussions in the context of women's lives and an impact on the management of the health system. It affects about 5-10% of women of reproductive age and clinically manifests with persistent pain (acyclic pain, dysmenorrhea, dyspareunia, dyschezia), although it is also associated with infertility. Laparoscopy makes the definitive diagnosis of the disease, although clinical treatment is endorsed with the presumed preoperative diagnosis. On the one hand, this approach avoids unnecessary procedures and speeds up treatment. On the other hand, it can trivialize the diagnosis with consequent overdiagnosis. But the biggest issue behind this paradigm is the lack of clinical criteria with good predictive capacity for endometriosis in the population of women with pelvic pain. Parallel to this, we have experienced a significant computational advance in recent decades. This has allowed considerable progress in interpreting large amounts of complex clinical data for example through the use of artificial intelligence. Based on this our aim is to explore modern machine learning techniques to train a preoperative prediction model of endometriosis based on clinical parameters. We will retrospectively analyze an anonymized database of 298 women aged between 18 and 50 years who underwent diagnostic laparoscopy at HCFMRP8USP for clinical suspicion of endometriosis. Data will be randomly sampled and stratified by surgical findings and separated into samples for training and validation at a ratio of 0.25. We intend to apply the following machine learning models: logistic regression support vector machine k neighbors classifier naive Bayes, decision tree classifier, random forest, gradient boosting classification, and multilayer perceptron (MLP). The project was approved by the local CEP on August 2, (Opinion No. 4,879,271, CAAE: 49523621.70000.5440). From a clinical-scientific point of view, we intend to develop a tool that helps healthcare professionals in the effective prediction of endometriosis with clinical data. Good preoperative selection of patients is essential, especially in this pandemic scenario with the need to optimize surgical indications. From the point of view of training human resources, we intend to train a student in the methodology and develop their critical thinking in the use of technology. (AU)