2787 - Automated Radiation Treatment Planning for Low-Dose Radiotherapy in Knee Osteoarthritis

Purpose/Objective(s):

Low-dose radiation therapy (LDRT) for osteoarthritis (OA) addresses growing demand for non-surgical management but adds operational burden to radiation departments primarily structured for cancer care. We developed an automated planning workflow for knee LDRT and hypothesized it would produce clinically acceptable plans comparable to manual planning.

Materials/Methods:

This single-institution, IRB-approved study generated automated LDRT plans (3 Gy in 6 fractions) on 21 previously treated knees with manual planning. The automated workflow incorporated isocenter placement, auto-contouring of the target, laterality-aware AP/PA beam generation, jaw and MLC shaping to anatomic landmarks, optimized field weighting, and 110% dose normalization. Four physicians independently scored both plan types using an institutionally developed rubric assessing joint coverage, hot spot control, and field setup (0–3 per domain; total 0–9). Scores were compared using paired Wilcoxon signed-rank tests. Standardized target contours were generated post hoc and applied uniformly to all plans to enable quantitative dosimetric comparison of joint coverage (PTV D95%, D90%) using paired t-tests. Inter-rater reliability was assessed via intraclass correlation coefficients (ICC).

Results:

Mean total scores were 7.9–8.3 (automated) versus 8.0–8.5 (manual), without statistically significant differences across readers (p>0.05). No automated or manual plans received a score of 0 in any domain. Acceptable scores (=2) were achieved in 100% (84/84) of manual versus 90% (76/84) of automated plans. Seven of eight borderline ratings (score 1) occurred in the field setup domain where there was inaccurate isocenter placement. Nevertheless, field setup scores were comparable across three readers (p=0.1); one reader favored manual plans (2.8±0.4 vs 2.5±0.5, p=0.03). Target coverage scores showed no significant differences across evaluators (p>0.05). Automated plans demonstrated more consistent hot spot control (2.95±0.22 vs 2.7±0.5, p>0.05) due to standardized normalization. Quantitative analysis demonstrated higher and more variable PTV coverage in manual plans (Table). Total score ICC was high for both workflows (0.91 manual, 0.86 automated).

Conclusion:

A novel automated planning system produced clinically acceptable knee OA plans that were comparable to manual planning across all scoring domains. Differences in quantitative coverage metrics identified an area for automation improvement. Automated planning has the potential to improve standardization and enable efficient integration of LDRT into busy radiation practices as treatment volumes for benign indications grow.

AI Statement: Claude 4.5 Sonnet and ChatGPT were used for language editing. Authors are responsible for all scientific content and conclusions.

PTV coverage	Manual	Automated	p-value
D90% (% prescription dose)	98.9 ± 3.6	92.8 ± 1.1	<0.001
D95% (% prescription dose)	91.1 ± 18.9	85.4 ± 2.4	0.002