Rationale and quality assessment of automated morphometry of the prostate gland on magnetic resonance imaging

Background: An AI service for automating prostate gland morphometry on MRI was tested, showing significant improvements over 11 existing systems. The real-time solution achieved 92% accuracy during prospective testing, highlighting its potential for use in radiology. Aim:Substantiate the necessity and evaluate the quality of automated measurements of the PG using magnetic resonance imaging (MRI) results in clinical healthcare practice Methods: The research had two stages: the first assessed the PG size in 3100 MRI protocols created by Moscow radiologists from 2022 to 2024 using the Data Set Preparation Platform tool. The second stage, from April to October 2024, utilized the STARD 2015 methodology to evaluate AI's role in automating measurements based on 560 MRI results and data from a morphometric AI service. Results:In the first stage, 3100 MRI examination protocols were analyzed, excluding post-treatment data. 47.2% were normal, 27.1% had three dimensions without volume, and 13.9% lacked dimensions or subjective characteristics, failing to meet standards. To address this, automated AI-based morphometry is proposed to expedite measurements and lessen the workload for doctors. In the second stage, the AI service's accuracy was assessed with 560 MRI scans, achieving 71.6% compliance. False negatives were 3.9%, and technical defects were 3.8%. The overall clinical assessment ranged from 88.0% to 97.0%, indicating high diagnostic quality, with main errors in gland contouring (52%) and incorrect measurements (13%), often due to prostate apex prolapse. Conclusions: Analysis of MRI protocols showed that 6.8% have low quality descriptions of prostate size, and 13.9% lack information about the size and volume. In 53.5% of cases, doctors do not confirm their conclusions with data. AI technologies for automating pancreatic measurements have a clinical score of 92.0%, which confirms their accuracy. The data can help in the development of new MRI-based automated morphometry products.