Determining the accuracy of an artificial intelligence-based model to automate the screening phase of related studies in the systematic reviews and comparing it with existing models in terms of strengths and weaknesses

Background and aims: Introduction: Systematic review and meta-analysis, which is often followedby it, form the cornerstone of evidence-based medicine. In recent years, artificial intelligencehas made great progress, so that today the daily life of humans is tied to this technology. .Since the process of conducting a systematic review is very time-consuming and expensive, it isnecessary to design a tool that can intelligently and automatically take over human tasks in thisprocess. Students of Tabriz University of Medical Sciences in an interdisciplinary team have succeededin designing a model based on artificial intelligence that can recognize articles related tothe subject of the study and is used in the screening stage of studies in the process of a systematicreview. The purpose of this study is to determine the accuracy of this system and compare it withother existing similar systems in terms of strengths and weaknesses.Method: Search terms were received from the co-researchers. After searching with each of theterms, the results were saved in a file and sent to fellow researchers. After categorizing the articlesby fellow researchers and the examined model, the results of the comparison and the accuracy ofthe model were calculated. Then, by referring to the BOX TOOL SR website, a search was madeto find similar models, and the characteristics of each of the tools were extracted and then withThe investigated model and other models were compared.Results: After comparing the results of screening studies by humans (collaborating researchers)and the case model, it was calculated that Precision = ۰.۸۳, Recall = ۰.۹۸, and Accuracy = ۰.۹۴.One of which is the investigated model.Conclusion: The reviewed model can help researchers with high accuracy without worryingabout missing relevant studies in the systematic review process, and save their time and energy.This model is also equal to the best available similar tools in terms of weaknesses and strengths.Improving it and developing a tool of this model that researchers can easily use will acceleratethe process of systematic review studies, which leads to the development of evidence-based medicine.Also, if it is possible to increase the precision of this model by maintaining a high recall, animportant step has been taken towards achieving a live systematic review.

کلیدواژه ها:

systematic review ، meta-analysis ، artificial intelligence ، automation ، title and abstract screening