Using Artificial Intelligence for Text Screening in a Systematic Review of Cardiotoxicity

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Systematic Review Methodology

Background:

• Traditional systematic reviews (SRs) involve time-consuming manual literature screening.
• Artificial intelligence (AI) offers potential for rapid data analysis and optimization of SR processes.
• The INSIDE platform was developed to support AI-driven decision-making in research.

Purpose of the Study:

• To compare the efficiency and quality of AI-based literature screening against traditional methods for systematic reviews.
• To assess the performance of the INSIDE platform in the context of prostate cancer research.
• To determine if AI can enhance the speed and accuracy of identifying relevant publications.

Main Methods:

• Comparative analysis of AI-based screening (INSIDE platform) versus traditional methods using four SRs on prostate cancer.
• Evaluation of efficiency using Work Saved Over Sampling (WSS) metrics at 80% and 95% relevant publication identification.
• Quality assessment through data visualization (scatter plots) to categorize records as relevant, irrelevant, or not screened.

Main Results:

• AI-based screening demonstrated higher efficiency, requiring fewer publications to identify key relevant records (WSS@80% 20.3%, WSS@95% 9.4%).
• Active learning within the AI approach further increased screening efficiency (WSS@80% 54.0%, WSS@95% 54.8%).
• Data visualization aided in broader search result analysis and identification of outlier articles.

Conclusions:

• AI-based approaches significantly optimize the systematic review process, enhancing efficiency and potentially quality.
• The study provides benchmarks for assessing AI tool performance in literature screening.
• Integration of AI into future systematic reviews is supported, with a need for continued testing and refinement.