Changes in CT Attenuation Values of Pericardium Months after Radiotherapy as Surrogates for Later Major Cardiotoxicity

Abstract

<h3>Purpose/Objective(s)</h3> Radiation-induced pericardial disease (RIPD) can manifest as pericardial swelling, thickening, effusion, fluid or calcification. While mostly asymptomatic or underdetected due to self-limiting nature, RIPD could be a surrogate marker for radiation-induced cardiotoxicity. Pericardial screening is not part of standard of care (SOC) for thoracic cancer patients' radiotherapy (RT). Here, we analyze SOC images in patients treated with thoracic RT for radiographically identifiable pericardial changes. We hypothesize that these changes have potential signatures for predicting late major cardiac events. <h3>Materials/Methods</h3> In this retrospective study, pericardium was manually contoured following standard contouring atlases in 30 locally advanced non-small cell lung cancer patients (15 females, 15 males) treated with chemoradiation. A 4mm inner wall was generated as the region of interest. Using pre-RT and (average 10-month) post-RT CT scans, we investigated whether changes in pericardium Hounsfield unit (HU) could predict major late cardiac events reported in patients' clinical records. Taking the effect of contrast-enhancement in some scans into account, HU value ranges representing effusion/fluid, normal tissue and calcification were considered as [-10,30], [31,132] and [133,500], respectively. Support vector machine analysis was performed using data from 24 patients with ≤12 months follow-up for training, and data from the remaining 6 patients for testing. Other parameters of interest for late cardiotoxicity were pericardial doses (mean and max), patient age at the time of RT, sex, history of cardiovascular diagnoses, smoking status, diabetes and hypertension. The CT scans were also revisited by a radiologist for verification of the existence of effusion, fluid, calcification or implants. <h3>Results</h3> In the training group, 11 patients, and in the testing group, 1 patient had a reported major cardiotoxicity several years post-RT. In the cohort of 30 patients, median changes, from pre-RT to post-RT, in the volumes of low, mid, and high HU regions were 17, 24 and 27cc, respectively. In the training set, the volumes of [31,132] HU region showed significant reduction (Mann-Whitney non-parametric test) from pre-RT to post-RT in CT scans of the patients with late cardiotoxicity (p=0.049) compared to those without. Also, patient age at the time of RT was a significant predictor of late cardiotoxicity (p=0.008), while mean dose to pericardium showed borderline significance (p = 0.09). Due to the confounding effect of both contrast-enhancement and implants on the higher HU value analysis, the accuracy of the classification model was limited in training group to 92% and in the test group to 67% (total 4 errors in patients with pre-existing calcifications). <h3>Conclusion</h3> Our preliminary study suggests that radiation-induced cardiotoxicity may be modelled based on patient demographic, dosimetric parameters, and detectable changes in pericardium sac HU volumes.