Abstract 210P
Background
Vascularization is known to be linked to tumour growth. We explored the potential of automated lung tumour vascularity assessment as risk factor for landmark survival.
Methods
131 CT scans from the NSCLC Radiogenomics dataset were evaluated. All patients underwent surgery, with or without other therapies. Primary lung tumour (PLT) is segmented by a radiologist. Vessels segmentation (arteries and veins) is performed by an in-house deep learning segmentation model. Vascularity features extracted for the PLT include: amount of blood vessel connections, % of the tumour surface connected to vessels (%area), and basic statistics on the surface area of the connections. A generalized linear model (GLM) to predict landmark survival was trained with these features, tumour volume, and clinical factors. Stepwise feature reduction was performed to arrive to the final model. We present the odds ratio (OR) as well as a combined risk score for the remaining independent predictors.
Results
Landmark survival was reached for 86/131 (66%) patients. GLM analysis showed 4 independent risk factors related to landmark survival: %area (OR= 1.69 for >2.5%), age (OR=1.57 for >70 years), gender (OR=2.10 for male) and radiation therapy (OR=1.51 for yes). Smoking status and tumor volume were not retained in the final model. We also showed that risk of dying before landmark survival increased with increased number risk factors (0%, 18%, 40%, 48% and 100% for 0, 1, 2, 3, 4 risk factors, respectively).
Conclusions
Lung tumour vascularity is an independent risk factor for landmark survival. We showed an approach were it can be used as additional prognostic factor to inform clinical decision making and therapy planning.
Clinical trial identification
Editorial acknowledgement
Fabio Bottari, PhD, of Radiomics for providing medical writing support in accordance with Good Publication Practice ( GPP 2022) guidelines.
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
F. Blistein, F. Belmans, S. Goffart, L. Libert, R. Narasimhan, A. Corsi: Financial Interests, Personal, Full or part-time Employment: Radiomics. W. Vos: Financial Interests, Personal, Leadership Role: Radiomics. M. Occhipinti: Financial Interests, Personal, Financially compensated role: Radiomics. All other authors have declared no conflicts of interest.
Resources from the same session
227P - Biomarkers for novel NK checkpoint inhibitor anti LLT1 antibody, ZM008: Patient transcriptome analysis
Presenter: M. S. Madhusudhan
Session: Poster session 01
228P - Biomarker discovery via meta-analysis of immunotherapy clinical trials in cancer
Presenter: Benjamin Haibe-Kains
Session: Poster session 01
230P - Fibroblast growth factor receptor (FGFR) co-alteration (co-alt) landscape and its impact on erdafitinib (erda) response in the phase II open-label, single-arm RAGNAR trial
Presenter: Yohann Loriot
Session: Poster session 01
231P - Time to deterioration of quality of life as a surrogate of OS: A trial-level surrogacy analysis
Presenter: Adel Shahnam
Session: Poster session 01
232P - Identification of ‘secondary’ germline cancer predisposition variants from somatic tumour testing: Should we evolve towards universal screening?
Presenter: Alejandro Gallego Martinez
Session: Poster session 01