Purpose: In this work we propose a novel methodology for assessing material differentiation in DECT that depends on the variance between measurements. Whereas traditional formalism describes detection with respect to a “blank” sample, we propose to extend the definition and generalize it so that we can produce quantitative results such as sensitivity, specificity, and area under the curve (AUC) for the differentiation of two materials in DECT.
Methods: To measure the total inter-measurement variance associated with a DECT measurement, we used the Gammex multi-energy CT phantom to take measurements of different material rod inserts at all locations within the phantom. Measurements were taken for five scans and 17 different slices over the 16 cm volumetric scan range for a total of 85 ROI measurements for each rod and for each DECT map. We also took DECT measurements of an off-centered phantom to account for variability in measurements due to an off-centered patient. We then used the mean and total inter-measurement variance to model the probability distribution function (PDF) of a single ROI measurement as a multivariate Gaussian distribution. Once the PDF is obtained, we can quantify both the degree and consistency with which two materials can be differentiated.
Results: When using only the iodine map, we tested this method on the 2.0 mg/mL iodine and blood + 2.0 mg/mL iodine rod inserts and obtained a classification AUC of 0.78, 0.80, 0.83, 0.84, and 0.85 when measured with an ROI of radius 5, 10, 15, 20, and 25 pixels, respectively. We also demonstrated that material differentiation between some materials can be improved using a multidimensional approach with more than one DECT image set.
Conclusion: Our proposed method allows for material differentiation using DECT to be measured quantitatively and generalizes easily to multi-dimensional data such as using multiple DECT image maps.