Imaging as Part of a Quality Assurance Program: Predictors of Interobserver Variability for Pretreatment Image Registration for Lung SBRT

PURPOSE: To evaluate the magnitude of interobserver variability in pretreatment image registration for lung stereotactic body radiation therapy patients in aggregate and within 3 clinical subgroups and to determine methods to identify patients expected to demonstrate larger variability. METHODS AND MATERIALS: Retrospective image registration was performed for the first and last treatment fraction for 10 lung stereotactic body radiation therapy patients by 16 individual observers (5 physicians, 6 physicists, and 5 therapists). Registration translation values were compared within and between subgroups overall and between the first and the last fractions. Four metrics were evaluated as possible predictors for large interobserver variability. RESULTS: The mean 3-dimensional displacement vector for all patients over all comparisons was 2.4 ± 1.8 mm. Three patients had mean 3-dimensional vector differences >3 mm. This cohort of patients showed a significant interfraction difference in variance (P value = .01), increasing from first fraction to last. A significant difference in interobserver variability was observed between physicians and physicists (P value < .01) and therapists and physicists (P value < .01) but not between physicians and therapists (P value = .07). Three of the 4 quantities evaluated as potential predictive metrics showed statistical correlation with increased interobserver variation, including target excursion and local target/lung contrast. CONCLUSION: Variability in pretreatment image guidance represents an important treatment consideration, particularly for stereotactic body radiation therapy, which employs small margins and a small number of treatment fractions. As a result of the data presented here, we have initiated weekly “registration rounds” to familiarize all staff physicians with the target and normal anatomy for each stereotactic body radiation therapy patient and minimize interobserver variations in image registration prior to treatment. The metrics shown here are capable of identifying patients for which large interobserver variations would be anticipated. These metrics may be used in the future to develop thresholds for additional interventions to mitigate registration variations.