University of Michigan Medical School Ann Arbor, MI
A. Tanweer1, M. P. Dykstra2, A. Hallstrom2, M. Mietzel2, J. R. Evans Jr2, S. R. Miller2, S. N. Regan2, S. Merkel2, S. Jolly2, M. M. Matuszak2, L. J. Pierce2, and R. T. Dess2; 1University of Michigan Medical School, Ann Arbor, MI, 2Department of Radiation Oncology, University of Michigan, Ann Arbor, MI
Purpose/Objective(s): Artificial Intelligence (AI) shows promise in streamlining clinical workflow in radiation oncology from initial consultation to treatment delivery. Stakeholder input is essential for this potential to be realized. Within the diverse practices of a statewide radiation oncology quality consortium, we sought to assess the current and future state of AI implementation in the radiation oncology clinic. Materials/
Methods: A structured online survey collected demographic information, clinical responsibilities, and workflows. In February 2024, the survey was distributed to the consortium membership using email and a QR code. The primary objective was determining AI use in Radiation Oncology as part of New Consultation, Care Navigation, Treatment Planning, and Plan Quality and Delivery. Secondary objectives included perceived barriers and facilitators to future AI adoption. Summary descriptive statistics were used; somewhat agree and strongly agree were combined to quantify responses. Results: Of 52 respondents, 40% were physicians (n = 21/52), 21% physicists (n = 11/52), 17% dosimetrists (n = 9/52), 6% radiation therapists (n = 3/52), and 15% administrators (n = 8/52). Average age was 47 with 19 years of experience; 27% were from academic centers (n = 14/52) and 73% were from community and free-standing practices (n = 38/52), with 21 unique centers represented. Only two respondents (5%, n = 2/44) reported AI use outside of Treatment Planning. Within Treatment Planning, however, normal-tissue AI contouring was reported by 51% (n = 22/43). Of those using or piloting automated contouring software (n = 26), 65% (n = 17/26) reported >1 hour of time savings per week and 38% (n = 10/26) endorsed >2 hours per week saved. Respondents agreed that AI tools have potential to increase consistency of normal tissue contouring (93%, n = 39/40), decrease contouring time (88%, n = 35/40), improve consistency of treatment plans (83%, n = 33), and decrease time from simulation to start (70%, n = 28/40). Most (82%, n = 32/39) agreed that they would like to use more automation for patients they treat. To justify implementation, 55% (n = 21/38) desired evidence of oncologic benefit or equivalence. Barriers to implementation included quality concerns (47%, n = 18/38) and administrative challenges regarding billing (50%, n = 19/38). Some (15%, n = 6/39) were concerned about automation threatening their jobs. Conclusion: Within a diverse statewide Radiation Oncology consortium, the predominant current AI use is concentrated within treatment planning where most respondents agree that AI can enhance workflow consistency and efficiency. The identified barriers to AI adoption highlight the need for further clinical validation, with a focus on quality along with administration and implementation support.
