Implementation of AI-based detection aid has no impact on radiologists’ workload, stress

UPDATED: Thursday, Dec. 14, at 11:30 a.m.

Implementation of an artificial intelligence-based detection aid appeared to have no impact on reducing radiologists’ workload or stress, according to recently published research.

Numerous studies have explored AI’s use in medical imaging. However, these investigations often do not take radiology’s complex work environment into account, researchers detailed Dec. 6 in the European Journal of Radiology [1].

German scientists recently sought to explore this issue, measuring how a computer-aided detection system for prostate MRI impacted radiologists’ workflow, caseload and stress.

“Our results do not support the broadly proposed expectation that AI leads to an increased efficiency in radiology in [the] form of decreased workflow-throughput times per case,” Katharina Wenderott, with the Institute for Patient Safety at University Hospital Bonn, Germany, and colleagues advised.

The CAD system in question was Quantib Prostate, a web-based, deep-learning MRI reading and reporting platform that is FDA cleared in the U.S. Wenderott and co-authors analyzed a total of 91 case reads, including 50 from prior to implementation and 41 afterward. The analysis unearthed variation in routine workflows after University Hospital started using the software. For instance, there was a nonsignificant increase in overall workflow throughput time stemming from CAD implementation (at 16.99 minutes vs. 18.77 after implementation).

Diagnostic reading times for high-suspicion cases increased after the institution started using CAD (15.73 minutes vs. 23.07). Meanwhile, radiologists did not report any changes to their workload or stress levels after they started using the technology.

“As our findings do not support the widely proposed assumptions that AI reduces radiologists’ workload, our study highlights the urgent need for high-quality research evaluating AI tools in routine clinical workflows,” the authors noted. “At the same time, carefully considering implementation issues around AI-facilitated technologies can improve integration of AI solutions into routine workflow and help to solve the pressing issues of radiologists’ workload and rising number of cases while safeguarding patient care.”

---

Update: Quantib issued the following response to the study, attributed to Lizette Heine, director of clinical science, prostate:   

"Previous publications, such as Günzel et al.'s study on Quantib Prostate AI (now Saige Prostate), emphasized clinical performance, revealing a 10% increase in prostate cancer identification and a 23% detection rate of additional cancerous locations (2022 ESUR).
 
The variations between past studies and the present report prompt consideration. Notably, this study centers on trainees rather than experienced radiologists, explores different trainee groups pre- and post-AI deployment, and focuses on timing, diverging from the software's primary cancer detection objective. A reexamination of technical logs exposes unintended deviations in IT hardware specifications.
 
Undoubtedly, the deployment methodology of AI warrants meticulous attention. We extend our commendations to the authors for their valuable contribution and look forward to working with them and others to bring the full promise of AI to patients."