AJR InBrief Banner

September 2015

Predicting Survival of Patients With Metastatic Melanoma Using CT Tissue Analysis

Andrew D. Smith

At the same time as former President Jimmy Carter announced that he is being treated for metastatic melanoma, new research published in the September issue of AJR describes a method of predicting overall survival of patients with the disease who were treated with a similar drug.

Coauthor Andrew D. Smith, associate professor of radiology at the University of Mississippi Medical Center, talked with InBrief about the study.

Why did you chose to investigate this particular patient population?

Melanoma metastases are considered highly vascular, similar to renal cell carcinoma (RCC) metastases. Our team has experience with combining serum factors with semiquantitative tumor imaging response evaluation methods (MASS Criteria) for predicting antiangiogenic response to metastatic melanoma. We also have experience with quantitative tumor imaging response evaluation methods for measuring changes in tumor heterogeneity (CT texture and histogram analysis) for predicting antiangiogenic response to metastatic RCC. Studying patients with metastatic melanoma allowed us to build upon our quantitative imaging knowledge and apply it to a different tumor.

What were the major findings of the study?

We found that a combination of changes in tumor CT texture and histogram analysis changes, tumor size changes, and a serum biomarker were able to accurately predict overall survival in patients with metastatic melanoma who had been treated with antiangiogenic therapy. The classic CT imaging biomarker for predicting tumor response to therapy is a change in tumor size. Our findings match those of other investigators, who suggest that quantifiable changes in tumor vascularity and heterogeneity on routine CT images are also predictors of patient survival, especially when combined with changes in tumor size and a serum biomarker.

How does CT texture and histogram analysis work?

Just as tumor length allows us to quantify changes in tumor size, CT texture and histogram analysis allow us to quantify changes in tumor vascularity and heterogeneity by characterizing image intensity patterns. There are several ways to measure tumor size—longest length, bidimensional size, maximal area, and volume—but there are hundreds of ways to measure tumor vascularity and heterogeneity using CT texture and histogram analysis.

The CT histogram features used in our study included mean, standard deviation, mean positive pixels (the mean of the pixels measuring >1 HU), and kurtosis and skewness (pointiness and symmetry of the pixel intensity distribution). CT texture includes the frequency of pixel intensities as well as a relationship of the intensities in 2D or 3D space (for example, entropy describes the irregularity or complexity of pixel intensities in space). There are hundreds of other CT texture and histogram features that can be measured on CT images.

Which CT texture and histogram analysis features are best for predicting response to antiangiogenic therapy?

The answer is not yet known. The toughest issue for researchers interested in CT texture and histogram analysis is to narrow the list of metrics to a few that are highly reproducible across different tumors, therapies, CT image platforms, patient populations, and imaging centers. Many factors affect image intensity, including the amount or rate of injection of IV contrast, cardiac output, kilovolts peak, field of view, pixel size, and slice thickness. More work is needed to validate a select few CT texture and histogram analysis parameters that can be used in the era of precision medicine.

How will tumor response assessment using standard CT imaging change in the future?

We will work on large-scale collaborative studies to refine image acquisition parameters, develop faster and more reliable ways to quantify changes in tumor size, and identify the best CT texture or histogram analysis parameters for tumor response evaluation to certain drug classes. In addition, we will move beyond structured reporting and perform structured images analysis with computer-assisted or algorithmic guidance throughout the image assessment process.