CBCT-Derived Radiosensitivity Marker Associated with Radiation Pneumonitis

Abstract

Dose escalation of non-small cell lung cancer is limited by increasing risk of severe adverse events, including radiation pneumonitis (RP). Recent studies have demonstrated a relationship between a cone beam computed tomography (CBCT) derived marker (CDM) and lung density changes on follow-up imaging. This study investigates the relationship between the CDM and symptomatic radiation pneumonitis in a new dataset. The CDM was defined as the proportion of normal lung tissue voxels receiving > 20 Gy that demonstrated an intensity increase beyond an expected noise threshold as previously described in [Radiother Oncol 2015;117:17-22]. CDMs were defined for both the tenth (CDM10) and twentieth (CDM20) fractions. CDMs were extracted for NSCLC patients treated definitively (>54 Gy) with fractionated (1.8-2 Gy/fraction) radiotherapy at the Princess Margaret Cancer Centre between 2011 and 2015. The exclusion criteria included patients without CBCT or dose objects available for analysis, patients receiving neoadjuvant radiotherapy, patients with previous radiation treatment in the thorax, and patients who had an in-thorax recurrence or were lost to follow-up within 1 year without evidence of pneumonitis. Images were registered into a common volume using an intensity-based deformable image registration algorithm in the Elastix toolbox. Image registration accuracy was manually checked. All other image analysis was implemented in Matlab 2013. RP was determined from prospective clinical records scored by the treating oncologists at follow-up visits as per CTCAE v 4.0, and reviewed retrospectively. Correlation of dosimetric parameters (mean lung dose (MLD), V20) and the CDMs to RP events was assessed by logistical regression. Fourteen percent of patients had image registration failure due to large deformations. CDMs were extracted in the remaining 84 patients. The prescription dose range was 54-74 Gy, mean lung dose range was 2.2-26 Gy and V20 range was 3.4–36%. Fifty-four percent of patients had concurrent chemotherapy with cisplatin or carboplatin and etoposide. Symptomatic RP (≥ Grade 2) occurred in 26.1% of patients of these patients. Univariate logistic regression of the CDM20 was significantly correlated with symptomatic RP (P < 0.01), while MLD, V20 and CDM10 were not significantly correlated. Multivariate logistical regression with CDM20 and V20 was significant for both (P < 0.05). This study suggests a relationship exists between symptomatic RP and a CBCT response marker initially derived to predict density changes in follow-up images. CBCT is regularly acquired as part of routine clinical care, and a CBCT derived marker may allow for modifications to treatment including reductions in dose, closer monitoring, and earlier initiation of steroid treatment. Future work is needed to improve the CBCT image quality and the deformable image registration accuracy.