S2314
Physics - Machine learning and AI algorithms
ESTRO 2026
driven manual plans (MP) for the same patient cohort were included. Data on PTV/OAR dosimetry and monitor units (MUs) were extracted. Parameters were grouped into clinically-relevant categories: PTV: Target Coverage (D95/D98), Conformity (CI), Homogeneity (HI), and Hot Spot (D2/V105%). OAR: Mean Dose, High- Dose Volume, Low-Dose Spill, and Hot Spot. A random-effects model (REML) was applied to compute pooled standardized mean differences (Hedges’ g) with 95% CIs under a paired design (r = 0.5). Results: A total of 147 records were identified, 82 full-text articles were assessed, and 45 studies (H&N: 13, Chest: 12, Abd&Pelvis: 15, Other: 5) were included for quantitative synthesis (Fig. 1). Overall, No PTV or OAR parameter showed significant inferiority of AP compared with MP. AP achieved comparable PTV metrics for target coverage (SMD = − 0.22 [ − 0.53, 0.09]), conformity (SMD = − 0.17 [ − 0.46, 0.12]), and homogeneity (SMD = − 0.20 [ − 0.29, 0.70]), as all 95% CIs crossed 0 (Fig. 2A). In contrast, AP demonstrated highly significant improvements in OAR sparing across all aggregated categories, including Mean Dose, High- Dose Volume, Low-Dose Spill, and Hot-Spot (all p < 0.01, Fig. 2A). The organ-wise subgroup analysis (Fig. 2B) confirmed this trend. Notably, the rectum exhibited a significant reduction in mean dose (SMD = − 1.06 [ − 1.76, − 0.37], I ² = 92%), and the bladder showed a similar decrease (SMD = − 0.56 [ − 0.85, − 0.27], I ² = 83%). Furthermore, AP resulted in comparable MUs (SMD = − 0.14 [ − 0.81, 0.53]). High between-study heterogeneity (I ² > 80%) was attributed to variations in AP system type and disease site.
people, processes, and technology. RTCs which combined strong technical integration with multidisciplinary engagement created an environment in which AI could be effectively implemented. Therefore, RTCs should invest in creating organizational readiness and transparent governance mechanisms instead of solely focusing on AI tools themselves. Recognizing and managing these dimensions can reduce resistance and accelerate the translation of AI innovation in clinical practice. References: Bardhan, I., Kohli, R., Oborn, E., Mishra, A., Tan, C. H., Tremblay, M. C., & Sarker, S. (2025). Human-centric information systems research on the digital future of healthcare. Information Systems Research, 36(1), 1– 20. Fiss, P. C. (2011). Building Better Causal Theories: A Fuzzy Set Approach to Typologies in Organization Research. Academy of Management Journal, 54(2), 393–420. Thompson, R. F., Valdes, G., Fuller, et al. (2018). Artificial intelligence in radiation oncology: A specialty-wide disruptive transformation? Radiotherapy and Oncology, 129(3), 421–426. Zapata, T., Muscat, N. A., Falkenbach, M., & Wismar, M. (2023). From Great Attrition to Great Attraction: Countering the Great Resignation of Health and Care Workers. Keywords: AI Implementation, Multi-Case Study Digital Poster 4210 Efficacy of radiotherapy auto-planning on dosimetric quality: A systematic review and meta- analysis Lin-Shan Chou 1 , Chih-Yuan Lin 2 , Yen-Jung Chen 1 , Ti- Hao Wang 3,4 1 Heavy Particles & Radiation Oncology, Taipei Veterans general hospital, Taipei, Taiwan. 2 Biomedical Imaging and Radiological Sciences, National Yang Ming Chiao Tung University, Taipei, Taiwan. 3 Medicine, China Medical University, Taichung, Taiwan. 4 Radiation Oncology, China Medical University Hospital, Taichung, Taiwan Purpose/Objective: Automated radiotherapy planning (AP) has been developed to enhance plan quality, reduce inter- planner variability, and improve workflow efficiency. However, reported outcomes vary across systems and disease sites. This study provides a systematic review and meta-analysis to quantitatively evaluate the dosimetric and efficiency outcomes of AP compared with manual planning (MP). Material/Methods: A systematic literature search of PubMed was performed to identify studies published from January 2020 to September 2025, per PRISMA guidelines. Studies comparing AP (KBP, MCO, etc.) with expert-
Fig. 2. Random-effects meta-analyses comparing dosimetric outcomes between automatic and manual planning.
Made with FlippingBook - Share PDF online