ABSTRACT
Radiology-structured reports (SR) have many advantages over free text (FT), but the wide implementation of SR is still lagging. A powerful tool such as GPT-4 can address this issue. We aim to employ a web-based reporting tool powered by GPT-4 capable of converting FT to SR and then evaluate its impact on reporting time and report quality. Thirty abdominopelvic CT scans were reported by two radiologists across two sessions (15 scans each): a control session using traditional reporting methods and an AI-assisted session employing a GPT-4-powered web application to structure free text into structured reports. For each radiologist, the output included 15 control finalized reports, 15 AI-assisted pre-edits, and 15 post-edit finalized reports. Reporting turnaround times were assessed, including total reporting time (TRT) and case reporting time (TATc). Quality assessments were conducted by two blinded radiologists. TRT and TATc have decreased with the use of the AI-assisted reporting tool, although statistically not significant (p-value > 0.05). Mean TATc for RAD-1 decreased from 00:20:08 to 00:16:30 (hours:minutes:seconds) and TRT decreased from 05:02:00 to 04:08:00. Mean TATc for RAD-2 decreased from 00:12:04 to 00:10:04 and TRT decreased from 03:01:00 to 02:31:00. Quality scores of the finalized reports with and without AI-assistance were comparable with no significant differences. Adjusting the AI-assisted TATc by removing the editing time showed statistically significant results compared to the control for both radiologists (p-value < 0.05). The AI-assisted reporting tool can generate SR while reducing TRT and TATc without sacrificing report quality. Editing time is a potential area for further improvement.
ABSTRACT
OBJECTIVES: To gain preliminary insight by exploring ulnar variance changes in a Saudi-based sample. METHODS: This 6-month (December 2013 to June 2014) cross-sectional study was conducted on a randomly selected healthy adult volunteers with a sample size of 104, at King Abdulaziz University Hospital, Jeddah, Saudi Arabia. Posteroanterior (PA), anteroposterior (AP), and PA grip views are taken. The variables of interest were the PA, AP, and PA fist measurements of both right and left wrists. An independent t-test was used to compare means between groups. RESULTS: A total of 104 volunteers were recruited. Among 17 participants who had a negative ulnar variance on right PA views, a significantly high proportion (n=9; 56.2%) maintained a negative value on fist views; 7 participants (43.8%) had a neutral ulnar variance while none (0%) had a positive value (p less than 0.001). Similarly, a significant proportion of participants who had neutral, or positive values on right PA views maintained the same values on right fist views (p less than 0.001). On radiographs of the right wrist, the ulnar variance decreased with a change in wrist position, with an absolute difference in magnitude of 2.13 (p less than 0.001) between PA and AP views. Similarly, the ulnar variance on the left side decreased significantly between PA and AP views (absolute difference in magnitude, 1.68; p less than 0.001). CONCLUSIONS: Ulnar variance changes in our sample are similar to what is reported in the literature.