Treatment for women diagnosed with ovarian cancer includes surgical resection of primary tumors and a chemotherapy regimen to treat metastasized tumors. Responses to chemotherapy varies widely, and because of this, extensive research is being conducted worldwide to identify better methods for imaging modalities to more accurately predict and monitor treatment responses.
A multi-specialty team at the University of Oklahoma campuses in Norman and Oklahoma City are investigating the role of applying quantitative imaging features computed from computed tomography (CT) for early prediction of tumor response to chemotherapy. Based on their most recent analysis of pre- and post-treatment CT scans, they report that the addition of quantitative imaging feature data helps improve the predictive accuracy of both imaging marker data computed from pre-therapy CT scans and radiologists’ assessments of Response Evaluation Criteria in Solid Tumors (RECIST) guidelines. The team’s promising early findings of predictive features have been published in Academic Radiology.
The 91 patients included in this study were participants in clinical trials at the university’s medical center. The patients had a diagnosis of recurrent, high-grade ovarian/peritonal/tubal carcinoma and had received systemic chemotherapy. They underwent a CT examination prior to chemotherapy treatment and a posttherapy CT scan performed 4-6 weeks after initiation of chemotherapy.
The objectives of this study were to investigate the potential utility of applying a new quantitative imaging feature analysis method and clinical marker acquired from pretherapy CT images or adding early posttherapy CT images to predict chemotherapy responses. The researchers also wanted to compare the prediction performance of the quantitative imaging markers they identified as most predictive with reported RECIST findings.
The researchers used a computer-aided detection (CAD) scheme they had previously developed to segment the tumors.1 This CAD scheme initially computed 159 features from the volumetric data of each segmented tumor in multiple CT image slices. The features were divided into four groups: a group of 10 tumor-shape-based features estimating tumor value and volume-based shape distortions, a group of 21 tumor density-based features, a group of two-dimensional grayscale-run length features, and a group of features based on wavelet transform. After the CAD scheme was applied to both pre- and post-therapy CT image data sets, the performance of each feature was evaluated and compared.
Lead author Gopichandh Danala and colleagues identified the 12 best-performed image features for both the pre-therapy CT images only and the difference of two CT image sets. These are described in detail in the article, but include skewness as the best performed tumor density feature, sphericity and compactness of tumor shape, and six wavelet features. By fusing a cluster of four selected optimal image features, predictive performance significantly increased. They reported that the new quantitative imaging marker generated from feature difference between pre- and post-therapy yielded higher prediction performance than using either imaging marker computed from pre-therapy images only or RCIST-guided assessment method (80.2% vs. 74.7%).
“The higher performance indicates that adding image features related to tumor shape (compactness) and density heterogeneity changes is helpful to more accurately predict tumor response to chemotherapy than using tumor size variation alone in the RECIST guidelines,” they wrote.
Additional investigation is under way. The authors did note that their study did identify a new quantitative imaging marker to predict clinical progression free survival at six months using pretherapy CT images alone. And by adding an early posttherapy CT scan, the team identified a second imaging marker based on the image feature difference between two sets of CT scans.
Applying quantitative CT image feature analysis to predict ovarian cancer treatment response. Appl Radiol.