Abstract P3-12-22: Reproducibility of automated coronary artery calcification scoring on radiotherapy treatment planning computed tomography scans of breast cancer patients

Abstract

Abstract Introduction: Presence of coronary artery calcifications (CAC) is a major independent risk factor for cardiovascular (CV) disease. CAC can be visualized on CT scans. Most breast cancer patients planned for radiotherapy (RT) receive planning CT scans. These scans may provide a reliable estimate of a patients' CV risk. This study evaluated the feasibility and reproducibility of an automated algorithm for CAC scoring on RT planning CT scans of breast cancer patients. Methods: This study was conducted within the Utrecht cohort for Multiple BReast cancer intErvention studies and Long-term evaLuAtion (UMBRELLA), and includes 562 breast cancer patients undergoing RT at University Medical Center Utrecht. Planning CT scans were performed using a 16-slice scanner (16 x 0.75 mm collimation, 3 mm thickness, 120 kVp, with or without breath hold (BH), without ECG synchronization). CAC were automatically scored using an algorithm developed with chest CT scans that considers lesions >130 Hounsfield units as CAC. CAC were identified using a supervised pattern recognition based on texture, size, and spatial features. To test validity of automated CAC scoring, manually scoring by an expert was performed in all scans with CAC (n = 80) and a random sample of scans without CAC (n = 83). Interscan reproducibility of automated CAC scoring was assessed in patients having two scans (n = 295). All scans with CAC score ≥ 1000 were manually checked and corrected if appropriate. Agatston calcification scores were analyzed continuously and categorically (0, 1-10, 11-100, 101-400, >400). Agreement and reliability for categories were determined with proportional agreement (%) and linearly weighted kappa. Reliability of Agatston scores were assessed with Intraclass correlation coefficients (ICC). Results: Of 562 patients, 129 (23%) patients had CAC scores > 0 with a mean of 93 (standard deviation: 166). Four patients had CAC scores ≥ 1000, which were erroneous and corrected. Of the 163 CT scans scored manually and automatically, 58 (36%) were performed with BH. Proportion of agreement was 79% (95% Confidence Interval (CI): 0.72-0.85) for all 163 scans: 88% (0.76-0.95) for 58 scans with BH and 74% (0.65-0.82) for 105 scans without. Proportion of agreement beyond chance was 0.80 (95% CI: 0.74-0.87) for all scans: 0.86 (0.77-0.96) with BH and 0.77 (0.684-0.853) without. Agatston score ICC was 0.86 (95% CI: 0.82-0.90) for all scans: 0.95 (0.91-0.97) with BH and 0.67 (0.55-0.76) without. For the interscan reproducibility (n = 295), the majority of patients (81%) had one scan with BH and one scan without. Proportion of agreement was 84% (95% CI: 0.79-0.88) and reliability was 0.61 (95% CI: 0.50-0.72). Agatston score ICC was 0.75 (95% CI: 0.69-0.80). Conclusion: Automated CAC scoring on RT planning CT scans of breast cancer patients is feasible. Agreement with manually scored scans is high and higher in CT scans performed with BH. Interscan reproducibility is fair. Automated CAC scores ≥ 1000 should to be manually checked and corrected if necessary. Automated CAC scoring on RT planning CT scans of breast cancer patients is available for all patients undergoing RT, and can provide information on CV risk at no additional cost. Citation Format: Gernaat SAM, de Vos BD, Isgum I, Rijnberg N, Bijlsma RM, Takx RAP, Pignol JP, Leiner T, Grobbee DE, van der Graaf Y, van den Bongard DHJG, Verkooijen HM. Reproducibility of automated coronary artery calcification scoring on radiotherapy treatment planning computed tomography scans of breast cancer patients. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P3-12-22.