《哈佛幻灯：CT影像组学预测非小细胞肺癌临床 结果》由会员分享，可在线阅读，更多相关《哈佛幻灯：CT影像组学预测非小细胞肺癌临床 结果（27页珍藏版）》请在金锄头文库上搜索。
1、CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller, Patrick Grossmann, Emmanuel Rios Velazquez, Raymond Mak, Hugo Aerts Dana Farber/ Brigham and Womens - Cancer Center Harvard Medical School Science Council Session The Physics of Cancers AAPM 56th annual meeting
2、, July 23rd 2014 CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Radiomics Medical imaging captures tumor phenotypic differences Radiomics quantify these differences using a large set of image features1 Quantitative radiomics data can be linked with clinical
3、and genomic information2 Currently only simple metrics based on tumor size are routinely quantified for clinical purpose 1. Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: Extracting more information from medical images using advanced feature analysis. Eur. J. Cancer. 2012;48(4):441446. 2
4、. Aerts HJWL, Velazquez ER, Leijenaar RTH, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat. Commun. 2014;5. CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Study design Aim 1.Ability of the radiomics to pre
5、dict distant metastasis 2.Distinguish better predictors than clinically used ones to detect patients at risk Data 182 NSCLC adenocarcinoma patients, overall stage 2 and 3 CT scan for radiotherapy treatment planning Primary tumors manually segmented and supervised by experienced radiation oncologists
6、 CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Radiomic Workflow CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Radiomic Workflow CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Corol
7、ler Radiomic Workflow CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Radiomic Workflow The clinical outcomes included : (not formed at the time of the planning CT) Survival From treatment start date to last date known alive Distant Metastasis (DM) Yes No Nod
8、al involvement N0 : node involved N+ : 1 or more nodes involved The analysis performed : Full dataset (n=182) split into: Discovery (n=98) dataset Validation (n=84) dataset Univariate and multivariate analysis CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller D
9、iscovery dataset heatmap CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller CI-Survival VS CI-Distant Metastasis CI - Metastasis CI - Survival CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Significant predictors (FDR co
10、rrected 0.05) Distant Metastasis (DM) 520 univariate predicting radiomic features Survival 582 univariate predicting radiomic features Strong significant predictors (FDR corrected 0.05 & C-index 60%) Distant Metastasis (DM) 35 univariate predicting radiomic features Survival 12 univariate predicting
11、 radiomic features Results Discovery dataset (n=98) CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Metastasis univariate analysis TOP 15 radiomic features Routinely used metrics CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud
12、Coroller Radiomic signature for Metastasis CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Clinical Model 1.Overall stage 2.Tumor grading CI = 0.57 P-value 9.6e-15 Multivariate models for DM CI = 0.61 P-value 4.6e-17 Combined Model 1.Overall stage 2.Tumor gra
13、ding 3.Radiomic predicting score Radiomic model 1. Radiomic signature CI = 0.62 P-value 7.85e-18 Models comparison CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Radiomic signature prediction score CT based radiomic features as predictors for clinical outcom
14、es of NSCLC Thibaud Coroller Nodes univariate analysis Ongoing analysis: Different cohorts including only 112 patients Overall stage 1, 2 and 3 No filter based features TOP 15 radiomic features Routinely used metrics CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Cor
15、oller Summary Analyzed primary lung tumor planning CTs No metastasis / nodes involvement at time of CT acquisition Successfully predicted future metastasis forming by radiomic signature Indication of successful future node involvement prediction (preliminary univariate analysis) Radiomics features h
16、ave better predicting power than clinically used features today CT based radiomic features as predictors for clinical outcomes of NSCLC Thibaud Coroller Patrick Grossmann, PhD Candidate Chintan Parmar, PhD Candidate Emmanuel Rios-Velazquez, PhD Hugo Aerts, PhD Ying Hou, MD Raymond Mak, MD Ralph Leijenaar, PhD Candidate Phillipe Lambin, MD, PhD Acknowledgment CT based radiomic features as predictors for clinical outcomes of NSCLC Thib
