SU‐E‐J‐79: Characterizing Accuracy in 4DCT Deformable Registration Using the POPI M odel

Abstract

Purpose: Deformable image registration can aid in fully utilizing 4DCT data without significantly increasing workload. In this study, we characterize the accuracy of a commercially available free‐form intensity‐based DIR algorithm on 4DCT using both successive registration through intermediate phases (SERIAL) and direct registration between the phases of interest (DIRECT). Methods: Two real‐patient 4DCT phantom datasets from the POPI model were used to test the SERIAL and DIRECT registration methods. One POPI model (patient1) displayed typical respiratory motion, 6.3mm +/− 3.0mm while the other (patient2) displayed significant respiratory motion, 14.0mm +/− 7.2mm, between end‐expiration and end‐inspiration. The phantoms were also used to study the relationship between initial rigid registration error and residual registration error after deformation. Results: For patient1, the residual errors after DIRECT and SERIAL deformation were 0.8mm +/− 0.4mm and 1.8mm +/− 1.4mm, respectively. For patient2, the errors after DIRECT and SERIAL were 5.1mm +/− 7.5mm and 1.6mm +/− 1.6mm, respectively. The correlation between initial and residual error was significant for both DIRECT (r = 0.72: linear, r = 0.80: second order polynomial) and SERIAL (r = 0.32: linear). The mean residual errors for points of interest with <5mm, 5–10mm, 10–15mm, 15–20mm, 20–25mm, and 25+mm initial error were 0.9mm, 0.9mm, 1.4mm, 4.7mm, 9.7mm, and 20.2mm after DIRECT and 1.3mm, 1.4mm, 2.1mm, 2.3mm, 1.9mm, and 3.1mm after SERIAL, respectively. Conclusion: DIRECT was more accurate for typical respiratory motion while SERIAL was more accurate when significant respiratory motion was present. There was flat correlation between residual and initial error for SERIAL as compared to a very steep correlation for DIRECT. Results show that the DIRECT approach delivers excellent results when the movement is limited to 15mm but when the movement increases, the SERIAL approach is favorable. Aaron Nelson and Jon Piper are both employees and owners of MIM Software Inc.