Click here to


Are you sure ?

Yes, do it No, cancel

Could Radiomic Signature Developed From NSCLC Patients Predict Overall Survival of Patients with ALK+ Mutation?

L Huang*, M Fan , W Hu , j wang , J Chen , X Xu , J Lu , G Qing , fudan university shanghai cancer center, Shanghai, shanghai


(Tuesday, 7/31/2018) 7:30 AM - 9:30 AM

Room: Davidson Ballroom B

Purpose: Radiomic features quantitatively represent tumor heterogeneity, which was widely investigated for predicting non-small cell lung cancer patients` outcome. But these studies focused on either NSCLC or its subgroup patients and did not mention their linkage. This study investigated the performance of a radiomics signature developed from NSCLC patients when it is directly applied to another subgroup patients who have ALK+ mutation and partly received targeted therapy.

Methods: Features with ICC value larger than 0.9 were considered as stable in test-retest. By using the LASSO Cox regression model and leave-one-out cross validation, the radiomic signature was developed and validated in NSCLC-Radiomic Collection. Then it was applied to test set to investigate for translational capability among NSCLC and its subgroups patients. C-index was calculated. Bootstrap was used to calculate 95% intervals. Rad_index was calculated as the linear summation of coefficients multiplied by feature value. Median Rad_index in training set was used as the threshold for stratification. And log-rank test was performed for stratified analysis.

Results: A radiomic signature based on five features (Geom_va_ratio, W_GLCM_LH_Std, W_GLCM_LH_DV, W_GLCM_HH_IM2 and W_his_HL_mean) was acquired. The C-index in training (0.632, 95% CI=0.630-0.634) and validation (0.621, 95% CI=0.616-0.626) set demonstrated a fair predictive power of this signature. In the test set, the overall C-index is 0.649 (95% CI=0.6401-0.658). And the signature has better performance in Non-Targeted therapy group (CI=0.842, 95% CI=0.832-0.852) than in Targeted therapy group (CI=0.573 95% CI=0.589-0.556). Moreover, only patients in Non-targeted therapy group can be correctly classified into high or low risk group (log-rank p-value = 0.00028).

Conclusion: This study for the first time demonstrated that radiomic signature from NSCLC could be applicable to patients with ALK+ mutation received Non-targeted therapy but not to those received Targeted therapy, which indicates that radiomic should be investigated within each subgroup of NSCLC to maximize its effectiveness.


Not Applicable / None Entered.


IM- CT: Quantitative imaging/analysis

Contact Email