J Christie¹*, O Daher¹, M Abdelrazek¹, P Lang¹, V Nair², S Mattonen¹, (1) Western University, London, ON, CA, (2) University of Washington School of Medicine, Seattle, WA

• 
  J Christie

• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  

SU-E-TRACK 6-1 (Sunday, 7/25/2021) 3:30 PM - 4:30 PM [Eastern Time (GMT-4)]

Purpose: To develop a multi-modal model integrating quantitative imaging features from the tumor and non-tumor regions of interest, descriptive features, and clinical data to improve risk-stratification of patients with non-small cell lung cancer (NSCLC) over cancer staging.

Methods: A dataset of 135 patients with early-stage NSCLC who had surgery as their primary form of treatment was analyzed. The tumor and peri-tumoral volumes on both CT and PET were segmented, while the bone marrow (L3-L5 vertebral bodies) was segmented on PET. The Quantitative Image Feature Engine was used to compute shape, size, first-order, and texture radiomic features from these volumes. Combined with the clinical and descriptive features, feature selection was performed on the 1030 features to select the top features to predict time to recurrence in the training cohort (n=94). A Cox model was built in the training cohort and evaluated on the testing cohort (n=41). Model performance was assessed using the concordance index and compared to a baseline clinical model with cancer staging.

Results: A total of 14 features were selected as the top performing features. Twelve of the top features were texture features with the remaining two features being cancer stage and patient age. The model achieved a concordance of 0.84 and 0.81 in the training and testing cohorts, respectively. The locked model outperformed that of a gold standard clinical model consisting of stage-only, with concordances of 0.69 and 0.68 in the training and testing cohorts, respectively.

Conclusion: Our model outperformed the stage-only model for NSCLC recurrence prediction and risk stratification. These results demonstrate that radiomics has the potential to augment staging information in a clinical setting. These non-invasive software tools can be implemented into the clinic and will allow physicians to more accurately identify patients who are at higher risk of treatment failure.

Funding Support, Disclosures, and Conflict of Interest: Natural Sciences and Engineering Research Council of Canada (NSERC)

Handouts

Keywords

Lung, Quantitative Imaging, CAD

Taxonomy

IM/TH- Image Analysis (Single Modality or Multi-Modality): Computer-aided decision support systems (detection, diagnosis, risk prediction, staging, treatment response assessment/monitoring, prognosis prediction)

Contact Email