Poster R12 - Assessing Pre-Treatment Risk of Cardiovascular Toxicity Among Cancer Patients Undergoing Chemotherapy using CMR Radiomics

Introduction: The advancement in cancer survivorship faces a potential setback as cardiovascular complications emerge as a prominent cause of non-cancer-related mortality among survivors. Anthracycline-induced cardiotoxicity (AIC), one such complication, is related to structural and functional changes in the heart, including damage to cardiomyocytes and fibrosis, and is reported to affect up to 34% of anthracycline-treated cancer survivors[1]. Current AIC detection via serial left ventricular ejection fraction (LVEF) measurements provides acceptable sensitivity only later in the disease progression[2]. It has been postulated that the harm caused by anthracyclines may balance or outweigh the potential benefits in some patients, necessitating means of identifying these patients at higher risk of adverse cardiovascular outcomes who might benefit from early cardioprotective interventions. Previous risk prediction models have used clinical variables (such as pre-treatment LVEF, body mass index (BMI), hypertension, age, etc.). However, these models are limited in many ways[3]. In this pilot study, our goal is to leverage pre-treatment tissue architecture information through radiomic texture features (RTF) to improve the performance of AIC risk prediction models.

Materials and

Methods: We performed this retrospective analysis of 65 patients undergoing anthracycline-based chemotherapy for stage I to III cancer (DETECT IV, R01CA167821). The participants underwent T1 mapping through cardiovascular magnetic resonance (CMR) imaging. LVEF assessment, patient demographics, and clinical data were also collected before the initiation of chemotherapy (baseline), at three months, and 24 months after the initiation of chemotherapy. The left ventricular (LV) myocardium was segmented from baseline T1 maps. Using pyradiomics, a set of 93 RTFs were extracted, each after preprocessing with a filter. RTFs, previously identified to be robust to changes in the radiomic workflow, were evaluated, and 5 RTFs were selected through recursive feature elimination for modeling. A comparator model incorporating established clinical variables as risk predictors (baseline LVEF, BMI, hypertension, age) was also developed [4]. The outcome was defined as a 10% decline in LVEF 24 months after anthracycline-based treatment and designated as a dichotomous variable, high risk (AIC) vs low risk (No AIC). For this pilot study, Various machine learning models were trained on clinical and RTF features using a five-fold cross-validation, and results were assessed using receiver operating curve (ROC) and area under the curve (AUC) metrics.

Results, Conclusions, and Discussions: Results and discussions: A total of 9 participants were excluded from the final analysis due to missing data, resulting in 56 patients (age: 54yrs ±14, 68% female and 32% male). Out of this, 17.8% (n=10) of patients met our criterion for high-risk AIC. The ROC curves in Figure 1 indicate that models based on RTFs achieved AUC values between 0.70 and 0.87, surpassing those built with clinical features (AUCs of 0.46 to 0.55). Our preliminary results show that pre-treatment LV myocardial RTFs possess a superior ability to predict patients at high-risk of developing AIC after anthracycline-based chemotherapy, suggesting that there may be markers in baseline tissue phenotypes that may be indicative of the patient’s susceptibility to AIC.

Conclusions: Using RTFs extracted from the LV myocardium in pre-treatment non-contrast T1 maps, we have been able to identify cancer patients at high-risk of cardiovascular toxicity caused by anthracycline-based chemotherapy in whom cardioprotective interventions should be considered. Stratifying patient's risk will aid clinicians in tailoring treatment plans to improve survivorship outcomes. Due to the small sample size employed in this study, a degree of caution is advised in interpreting the results. Future research efforts will be aimed at extending this study to larger and more diverse cohorts to validate our findings

Acknowledgements (Optional): REFERENCES
[1] M. Guglin, G. Hartlage, C. Reynolds, R. Chen, and V. Patel, “Trastuzumab-Induced Cardiomyopathy: Not as Benign as it Looks? A Retrospective Study,” J Card Fail, vol. 15, no. 8, 2009, doi: 10.1016/j.cardfail.2009.04.011.
[2] D. Cardinale et al., “Early detection of anthracycline cardiotoxicity and improvement with heart failure therapy,” Circulation, vol. 131, no. 22, 2015, doi: 10.1161/CIRCULATIONAHA.114.013777.
[3] E. G. Kaboré et al., “Risk Prediction Models for Cardiotoxicity of Chemotherapy among Patients with Breast Cancer: A Systematic Review,” JAMA Netw Open, vol. 6, no. 2, 2023, doi: 10.1001/jamanetworkopen.2023.0569.
[4] J. N. Upshaw et al., “Personalized Decision Making in Early Stage Breast Cancer: Applying Clinical Prediction Models for Anthracycline Cardiotoxicity and Breast Cancer Mortality Demonstrates Substantial Heterogeneity of Benefit-Harm Trade-off,” Clin Breast Cancer, vol. 19, no. 4, 2019, doi: 10.1016/j.clbc.2019.04.012.