Radiation Physics

PV QA 3 - Poster Viewing Q&A 3

TU_7_3182 - Time stability of delta radiomics features extracted from longitudinal CTs

Tuesday, October 23
1:00 PM - 2:30 PM
Location: Innovation Hub, Exhibit Hall 3

Time stability of delta radiomics features extracted from longitudinal CTs
T. E. Plautz1, G. Noid1, D. Schott2, C. Zheng3, and A. Li1; 1Medical College of Wisconsin, Milwaukee, WI, 2Department of Radiation Oncology, Medical College of Wisconsin, Milwaukee, WI, 3University of Wisconsin-Milwaukee, Milwaukee, WI

Purpose/Objective(s): We have previously reported on the feasibility of using radiation-induced CT texture changes (delta radiomics) measured from daily CTs during radiotherapy (RT) as an imaging biomarker for RT response in various tumors. This study aims to evaluate time stability in texture features over a timescale similar to the duration of conventional RT delivery.

Materials/Methods: CT data sets of a 3-D printed anatomically-informed texture phantom containing three modules (liver, lung and uniform with low-contrast lesions) and the homogeneous module of a quality assurance phantom were acquired at regular intervals over the course of an 8 week period, to simulate the timescale of conventional RT duration. A CT scanner installed in the RT room and our institution’s standard abdominal protocol (120 kVp, 126 mAs, 0.6 pitch, 1.2mm focal-spot size, 17 mGy CTDIvol, 3mm slice thickness, 0.988/0.988mm pixel spacing and B30f reconstruction kernel) were used. In each phantom module, 12 regions of interest (ROI) each with a volume of 8 cm3 were selected and 60 texture features including histogram, GLCM and GLRM features, were extracted using an in-house program. Each ROI was transferred to all images in the module series using rigid registration of the images. The time stability of each feature was evaluated by fitting the longitudinal datasets using linear mixed effects models. A 95% confidence interval (CI) was established to quantify the expected variation over a treatment timescale for each of the 60 features evaluated in each phantom. These results were compared with delta radiomics of first order textural features in pancreatic cancer RT patient data previously reported. It will also be used to evaluate the utility and practicality of higher-order textural feature analysis using current pancreatic cancer RT patient data.

Results: In the homogeneous module, the 95% CI for the variation in mean CT number (MCTN) is ±1.0 HU, i.e. variation in MCTN greater than 1 HU is significant at the p≤0.05 level. The expected variation in MCTN increased with increasing complexity of the phantom (1.5 HU, 2.1 HU and 8.3 HU for the low contrast, liver and lung modules respectively) which is primarily a result of the partial volume effect, due to the uncertainty in the position of the z-slices. In the homogeneous phantom, skewness varied by ±0.12 and kurtosis varied by ±0.25. These values confirm that the previously reported values of change in MCTN, skewness and kurtosis in patient daily CTs for pancreatic cancer of 4.7±4.5 HU, +0.12 and -0.39, respectively, are significant at the p≤0.05 level, with respect to the expected variation.

Conclusion: Machine-related uncertainty in CT texture is predictable. CT texture measurements of anatomically-informed texture phantoms are highly repeatable and stable. This indicates that previously reported radiation-induced changes in patients’ CT texture features are outside the range of expected variation, suggesting their potential for early assessment of treatment response.

Author Disclosure: T.E. Plautz: None. G. Noid: None. D. Schott: None. C. Zheng: None. A. Li: None.

Tia Plautz, PhD

Presentation(s):

Send Email for Tia Plautz


Assets

TU_7_3182 - Time stability of delta radiomics features extracted from longitudinal CTs



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Time stability of delta radiomics features extracted from longitudinal CTs