Room: Exhibit Hall
Purpose: To investigate the best machine learning method of non-small cell lung cancer for overall survival time prediction and test the association between Radiomics features and the overall survival time.
Methods: A total of 215 Radiomics features were extracted from the segmented tumor volumes of pretreatment CT images. These Radiomics features quantify tumor phenotypic characteristics on medical images using tumor shape and size, intensity statistics, and texture. For univariate analysis, the correlation coefficient was performed to assess each Radiomics featureâ€™s association with the overall survival time. For multivariate analysis, the performance of 5 feature selection methods and 5 regression methods were investigated for overall survival time prediction. The predicted performance was evaluated with concordance index between predicted and true overall survival time for 246 non-small cell lung cancer patients.
Results: In univariate analysis, the Radiomics features got from LLL processing CT image showed larger concordance index (CI) with OS. In multivariate analysis, random forest method with 10 features and Spearmanâ€™s correlation coefficient obtained the highest concordance index (0.68) compared to other regression methods.
Conclusion: The preliminary results demonstrated that certain machine learning and Radiomics analysis method could predict overall survival time of non-small cell lung cancer accuracy. Furthermore, it could be observed that Radiomics features shown significant association with the overall survival time of non-small cell lung cancer patient.