The effect of registration surrogate and patient factors on the interobserver variability of electronic portal image guidance during prostate radiotherapy

Abstract

Intraprostatic fiducial markers (IPM) and electronic portal imaging (EPI) are commonly used to identify and correct for prostate motion during radiotherapy. However, little data is available on the precision of this image-guidance technique. This study quantified impact of different registration surrogates and patient factors on the interobserver variability of manual EPI alignment during prostate radiotherapy. For 50 prostate radiotherapy patients previously implanted with 3 IPM, five observers manually aligned 150 pairs of orthogonal EPI to the reference digital reconstructed radiograph using Varian Vision EPI analysis software. Images were aligned using: Bony anatomy (BA), single mid-prostate IPM (SM); and 2 strategies using 3 IPM: center of mass (COM) and rotate & translate (R&T). Intraclass correlation coefficients (ICCs) were calculated to quantify interobserver variability. The absolute displacements measured using SM and R&T were compared with those using COM. The impact of patients' pelvic diameter and adjuvant hormone therapy on interobserver variability were also evaluated. Twelve thousand displacement values were collected for analysis. The maximum discrepancy between the 5 observers was >2 mm in 47% of measurements using BA, 5% using SM, 4% using R&T, and 3% using COM. Both of the 3 IPM alignment strategies demonstrated lower interobserver variability than the single IPM strategy (ICC 0.94-0.97 vs. 0.82-0.94). BA had the highest interobserver variability (ICC = 0.43-0.90). Pelvic diameter and hormone therapy had no discernible impact on interobserver variability. Compared with COM, the absolute displacements measured using the other IPM strategies were statistically different (p < 0.001), but 95% of the absolute magnitude of differences between the strategies were ≤1 mm. The high reproducibility among the observers demonstrated the precision of prostate localization using multiple IPM and EPI, which was not influenced by the patient factors studied. Bony anatomy displayed the highest interobserver variability of the 4 alignment, likely because of the limited EPI field-of-view. Alignment using more than one IPM is recommended to minimize interobserver variability.