Click here to


Are you sure ?

Yes, do it No, cancel

Radiomics for Early Detection of Recurrence in Patients Treated with Lung Stereotactic Ablative Radiotherapy (SABR)

T Kunkyab12*, B Mou 1, A Jirasek2, C Haston 2, J Andrews 2, S Thomas 3, D Hyde1, (1) BC Cancer Kelowna, Kelowna, BC, CA, (2) Department of Mathematics, Statistics, Physics & Computer Science, University of British Columbia Okanagan, Kelowna, BC, CA, (3) BC Cancer Vancouver, Vancouver, BC, CA


(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: Radiation-induced lung injury or fibrosis obscures follow-up imaging after SABR and can prevent the early detection of local recurrence. We investigated using radiomics to differentiate local recurrence from fibrosis.

Methods: We included post-treatment computed tomography (CT) scans (n = 168) of 79 patients treated with lung SABR. The training set consisted of 24 months median follow up with 16 recurrences and 31 non-recurrence (47 patients in total). Independent validation sets were separated into three cohorts by follow up CT scans at 24 months (cohort 1), 9 - 12 months (cohort 2), and 5 – 9 months (cohort 3). Cohort 1 contained 32 patients (22 non-recurrences and 10 recurrences), cohort 2 contained 27 patients (18 non-recurrences and 9 recurrences) and cohort 3 contained 30 patients (20 non-recurrences and 10 recurrences). All recurrences were identified by an experienced Radiation Oncologist using validated high risk radiologic features. We used leave-one-out cross validation, at each fold, 1-nearest neighbor with stepwise forward feature selection was used to create a matrix of most frequently selected features. The top five features were evaluated using Support Vector Machine classifier. PyRadiomics was used for feature extraction. Python and R were used for feature selection and modeling.

Results: The top five features selected from the training set were Dependence Non-Uniformity Normalized, Small Dependence Low Gray Level Emphasis, Run Length Non-Uniformity Normalized, Dependence Non-Uniformity and Small Dependence Emphasis. For validation cohort 1, the sensitivity was 70% and specificity was 100% (AUC = 0.85); for cohort 2, the sensitivity was 55.6% and specificity was 94.4% (AUC = 0.75); and for cohort 3, the sensitivity was 70% and specificity was 90% (AUC = 0.80).

Conclusion: Radiomic analysis may distinguish recurrences from non-recurrences as early as 5-9 months post-treatment. Three independent validation sets were used to determine the generalizability of this radiomic modeling.

Funding Support, Disclosures, and Conflict of Interest: Moss Rock Park Foundation Eminence Fund


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email