Purpose: Early prediction of tumor response is the key for plan adaption towards individualized treatment management. We aimed to investigate the value of longitudinal radiomics analysis of daily CBCT images for earlier prediction of neoadjuvant chemoradiation treatment (nCRT) response in patients with locally advanced rectal cancer (LARC).
Methods: A subset of 86 LARC patients who underwent nCRT with 50.4 Gy in 28 fractions was analyzed retrospectively. Clinical response scores according to Tumor Regression Grade (TRG), assessed from post-operative pathology, were collected for all patients. Among 19 cases stuided, 10 patients achieved pathological complete response (pCR), 9 patients achieved partial or no response (non-pCR). Daily CBCTs were imported to Varian Ethos emulator, where segmentation of the gross tumor volume (GTV) and organ-at-risks (OARs) were automatically generated. The delivered doses based on daily patient anatomy were calculated from the CBCT of the day on Varian Ethos emulator. For each fraction, we extracted 1304 imaging features from the GTV on CBCT using PyRadiomics package. Highly correlated imaging features were eliminated according to Spearman's rank coefficient. Longitudinal correlations between the selected imaging features, dosimetric parameters and the pathological response were established using linear regression models.
Results: Among 252 CBCTs assessed, the variations of the GTV and PTV coverage ranged 2.8% and 7.2% respectively over the course of the treatment. Dosimetric variations showed no clear relationships between pCR vs non-pCR patients. Correlations were observed using a small subset of imaging features and early treatment response detection as early as the 8th fraction.
Conclusion: Early prediction of treatment response is possible via systematical analyses of CBCT radiomics over the treatment course. Quantitative imaging features could be used as biomarkers to predict treatment response for LARC during the treatment course. Varian Ethos emulator offers an integrated platform to streamline the process.
Funding Support, Disclosures, and Conflict of Interest: The work is partially supported by Varian Inc.
Not Applicable / None Entered.