MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

A General Framework of Delta-Radiomics for Treatment Response Prediction

H Nasief*, X Li , Medical College of Wisconsin, Milwaukee, WI

Presentations

(Tuesday, 7/16/2019) 7:30 AM - 9:30 AM

Room: Stars at Night Ballroom 2-3

Purpose: Delta-radiomics assess the net change of radiomic features obtained from longitudinal images over the course of treatment and thus can be developed as an image biomarker for early prediction of treatment response. This study aims to develop a general framework for delta-radiomics.

Methods: The framework includes: (1) segmenting and registering longitudinal image set, (2) exacting and calculating delta-radiomic features (DRFs), (3) identifying DRFs with significant changes that correlates with outcome using T-test, regression models, and linear mixed effect model, (4) determining effective combination(s) of DRFs for outcome prediction using machine learning, and (5) validating the prediction model using a leave-one-out method and AUC under ROC curve. This process was tested using daily non-contrast CTs, acquired during CT-guided preoperative chemo-radiation therapy for 70 pancreatic head patients, along with their pathological response data. All patients were treated with 50.4Gy in 28 daily fractions. Over 1300 radiomic features were extracted from the segmented tumor regions. The effect of intra-fraction motion was assessed using coefficient of variance (COV) and the effect of acquisition parameters analyzed with a multi variate regression method.

Results: The results showed that 35% of the features were affected by the motions with COV>5%, of which 45% has a COV <10%. No significant correlations were found between DRFs and the acquisition parameters used. Spearman correlations showed that 73 radiomic features can be used for delta radiomic analysis, while 13 DRFs passed t-test and linear mixed effect (p<0.05) and demonstrated significant changes by 2-4 weeks of treatment. The best performing three feature combination (AUC=0.94) was obtained using Kurtosis, NESTD and Coarseness.

Conclusion: We have developed a general framework for delta-radiomic based on statistical methods and machine learning algorithms. The framework can be used to identify appropriate DRFs to predict treatment response as demonstrated for pancreatic cancer.

Keywords

Not Applicable / None Entered.

Taxonomy

TH- response assessment : CT imaging-based

Contact Email