Assessment of Image Registration Techniques during Lung Re-irradiation

Summary: Thoracic re-irradiation is increasingly used in patients with lung cancer, but anatomical distortions between radiation treatments from lung fibrosis and varying motion management techniques lead to uncertainties with image registration during re-irradiation planning. Optimizing image registration strategies may help clinicians more accurately determine cumulative radiotherapy doses and assess toxicity risk to sensitive normal tissues. Twenty-five patients were analyzed retrospectively who had undergone at least two instances of thoracic irradiation to the same lung. At least one course of SBRT was used in 22/25 patients, conventional fractionation in 10/25 patients, and palliative radiotherapy in 4/25 patients. The most common regimens were 48 Gy in 4 fx, 66 Gy in 33 fx, and 30 Gy in 10 fx respectively. Using an initial and subsequent treatment planning CT, a rigid image registration (RIR) was created based on spine and carina anatomy for each patient (using MIM software Inc.). Next, imaged-based deformable registration (DIR) was performed based on each RIR. Finally, a hybrid deformable registration using lung lobe contours (DIR w/ lobes) was performed from each RIR. Geometric differences between registrations were calculated based on pre-determined landmarks in the treated lung (29-30 landmarks per patient). Registrations that resulted in the smallest mean geometric difference between landmarks were considered to be the most accurate. RIR was the most accurate registration in 1/25 (4%) patients, and deformable registrations (both DIR and DIR w/ lobes) in 24/25 (96%) patients. DIR was the most accurate in 7/25 (28%) patients, while DIR w/ lobes was most accurate in 17/25 (68%) patients. Of these 17 registrations, 12/17 (70.6%) were carina-based while 5/17 (29.4%) were spine-based. Amongst carina registrations, there was a statistically significant difference between DIR and DIR w/ lobes (p=0.024), between RIR and DIR (p< 0.001), and between RIR and DIR w/ lobes (p< 0.001). Carina DIR w/ lobes had the lowest mean geometrical difference (6.4 mm). Amongst spine registrations, there was a statistically significant difference between RIR and DIR (p< 0.001) and between RIR and DIR w/ lobes (p< 0.001). The difference between spine DIR and DIR w/ lobes was not statistically significant (p=0.087). DIR w/ lobes had the lowest mean geometrical difference (6.6 mm). There was not a statistically significant difference between carina DIR w/ lobes and spine DIR w/ lobes (p=0.2). In our sample of 25 patients, hybrid deformable image registrations using lung lobe contours, based on carina anatomy rigid registrations, resulted in the smallest mean geometrical error. This was not statistically different from spine-based hybrid registrations. Although larger studies are needed for validation, our findings emphasize the importance of using lung lobe contours during deformable registration in patients undergoing thoracic re-irradiation. This may also lead to more accurate cumulative dose calculations during radiotherapy planning. Navid Fallahi, MD, MPH (Presenting Author) - Massey Comprehensive Cancer Center at VCU; Tianjun Ma, PhD (Co-Author) - Massey Comprehensive Cancer Center at VCU; Christopher Guy, PhD (Co-Author) - Massey Comprehensive Cancer Center at VCU; David Cooper, MD (Co-Author) - Tennessee Oncology; Nuzhat Jan, MD (Co-Author) - Massey Comprehensive Cancer Center at VCU; Elisabeth Weiss, MD (Co-Author) - Massey Comprehensive Cancer Center at VCU