The Diagnostic values of Machin learning models in early detectionof Endometrial Cancer: A Diagnostic Meta-Analysis

Objective: Endometrial cancer is considered one of the six most common types of cancer and thefourteenth leading cause of women’s cancer deaths worldwide. Due to the high number of deaths,Incidence, prevalence, and ultimately disease burden, early detection of this type of cancer isvery important for health and clinical decisions. This study was conducted to compare differentMachine-Learning (ML) models in predicting the occurrence, recurrence, and lymph node involvementof endometrial carcinoma and ultimately finding the best model for early detection ofthis type of cancer.Methods: This meta-analysis was gathered based on the PRISMA guidelines and the PIRTstructure. The preferred databases for searching included PubMed (Medline), Scopus, Web ofSciences, Embase, and Cochrane Library, from January ۲۰۰۰ to March ۲۰۲۳. Some of the basickeywords were “Endometrial Cancer”, “Machine Learning”, and the synonyms of these keywordswere retrieved through Mesh and Emtree. After the search, articles were screened based on thetitle, abstract and full text. And finally, using the researcher’s own checklist, data extraction wasdone. Finally, the quality assessment of the articles was accomplished based on the QUADAS-۲checklist, and data analysis was done with version ۱۷ of STATA software.Results: After the search, a total of ۱۸۳ articles were retrieved from international databases. Afterscreening based on title, abstract, and full text, ۶ studies met the inclusion criteria and wereconsidered for meta-analysis. Out of these ۶ studies, ۲ studies examined the diagnostic value ofdifferent ML models in the early detection of occurrences, ۲ studies in the early detection of recurrence,and ۲ other studies in the early detection of metastasis to lymph nodes in endometrialcancer. The pooled sensitivity, specificity, accuracy, AUC, PPV, and NPV of ML in general in thediagnosis of endometrial cancer was ۰.۸۰ with a confidence interval of (۹۵% CI: ۰.۸۳ - ۰.۷۷), ۰.۸۱(۹۵% CI: ۰.۷۸ - ۰.۸۴), ۰.۸۱ (۹۵% CI: ۰.۷۸ - ۰.۸۴), ۰.۸۵ (۹۵% CI: ۰.۸۳ - ۰.۸۸), ۰.۵۱ (۹۵% CI: ۰.۴۶- ۰.۵۶) and ۰.۹۶ (۹۵% CI: ۰.۹۵-۰.۹۶), respectively. ADC model with a sensitivity of ۰.۸۶ (۹۵% CI:۰.۵۷ - ۱.۰۰) was the most sensitive in diagnosing endometrial carcinoma. The highest AUC (۸۸%)was related to GBDT, NN, and RF models.Conclusion:The results of the present meta-analysis confirmed that different ML algorithms canbe beneficial in the early diagnosis of endometrial cancer.