Modern methods of computer interpretation of abdominal radiography: Experience of application in diagnostics

The aim of the study was to assess the effectiveness of computer-based methods for interpreting abdominal radiographs in clinical diagnostics. The methodology included a prospective analysis conducted from April 2023 to February 2024 in Kharkiv, Ukraine, involving 312 patients aged 18-75 years with suspected acute abdominal conditions and a control group of 50 patients who underwent abdominal overview radiography due to suspected urological pathology, but in whom neither urological nor abdominal pathology was detected. Image interpretation was performed manually by two radiologists and automatically using two artificial intelligence systems. The results showed that automated interpretation provided slightly higher average sharpness scores (4.7 ± 0.3 vs 4.6 ± 0.4) and contrast (4.6 ± 0.4 vs 4.5 ± 0.5) compared to manual evaluation, as well as fewer artefacts (4.5 ± 0.5 vs 4.2 ± 0.6). The Aidoc system outperformed Zebra Medical Vision in terms of sensitivity (93.6% vs 89.1%), specificity (95.4% vs 94.7%), positive predictive value (91.8% vs 88.2%), and negative predictive value (96.7% vs 92.5%). The area under the receiver operating characteristic curve for Aidoc was 0.972, compared to 0.951 for Zebra Medical Vision. Kappa coefficients indicated higher consistency of Aidoc with expert assessments in diagnosing bowel obstruction (κ = 0.92 vs 0.88) and pneumoperitoneum (κ = 0.91 vs 0.85). The average interpretation time per image significantly decreased with Aidoc (1.4 ± 0.3 minutes) compared to manual analysis (6.8 ± 1.2 minutes) and Zebra Medical Vision (1.9 ± 0.4 minutes). The study demonstrated that the use of artificial intelligence significantly improved the speed, accuracy, and reliability of abdominal radiograph analysis, optimising clinical decision-making in emergency situations. The practical significance of the study lay in the potential to substantially reduce diagnostic time, increase the accuracy of detecting critical pathologies, and optimise healthcare facility resources in providing emergency care