The Potential Role of Radiomic-Driven Treatment Decision-Making for Locally Advanced NPC
X Han1, X Teng1, J Zhang1, Z Ma1*, S Lam1, H Xiao1, C Liu1, W Li1, Y Huang1, F Lee2, J Cai1, (1) The Hong Kong Polytechnic University, Hong Kong, (2) Queen Elizabeth Hospital, Hong Kong
Presentations
WE-C1000-IePD-F1-4 (Wednesday, 7/13/2022) 10:00 AM - 10:30 AM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 1
Purpose: Concurrent chemoradiation therapy (CCRT), adjuvant chemotherapy (ACT) after CCRT, and induction chemotherapy (ICT) before CCRT are three mainstream treatment strategies for locally advanced nasopharyngeal carcinoma (NPC). However, patient-specific decision-making is still under debate. This study investigated the potential role of personalized treatment decision-making driven by radiomics in NPC patients by showing that the predictive radiomic features in predicting three-year progression-free survival (PFS) are treatment-specific and not generalizable between treatments.
Methods: A total of 299 NPC patients who received CCRT or ACT or ICT in local hospitals, were screened. Radiomics features were extracted from primary tumor volume on contrast-enhanced computed tomography images, and the low robust features were removed. The radiomic feature's generalizability was evaluated by 1) identifying the predictive radiomic features in a treatment group and 2) verifying whether these features are still predictive in other treatment groups. The log-rank test and univariable Cox model were used to identify the predictive features. The generalizability of predictive features is observed by comparing the log-rank test p-value between treatment strategies.
Results: This study identified 74 predictive features in the CCRT group for 3-year PFS stratification (p-value < 0.01), 14 predictive features in the ACT group (p-value < 0.01), and 3 in the ICT group (p-value < 0.01). Only 7.14% (1 / 14) features in the CCRT group are generalizable to other groups (p-value < 0.05). 14.3% (2 / 14) (p<0.05) features in ACT group are generalizable to other groups (p-value < 0.05). Furthermore, none of the predictive features in the ICT group are generalizable to other groups (p-value <0.05).
Conclusion: The CT radiomic features predicting 3-year PFS for locally advanced NPC patients are likely to be treatment-specific. An integrated evaluation of radiomic-based predictive models for different treatments is warranted for building an actual treatment-specific model to guide the treatment decision-making process.
Funding Support, Disclosures, and Conflict of Interest: ITS/080/19, Project of Strategic Importance (P0035421), The Hong Kong Polytechnic University and Shenzhen-Hong Kong-Macau S&T Program (Category C) (SGDX20201103095002019), Shenzhen Basic Research Program (R2021A067)
Keywords
Image Analysis, Image Guidance, Quantitative Imaging
Taxonomy
Contact Email