Automated calcium burden measurement in internal carotid artery plaque with CT: a hierarchical adaptive approach

Calcium burden measurement in internal carotid artery (ICA) plaque could play an important role in assessing stroke risk and stenosis quantification in the ICA. We propose an automatic method for labelling calcified plaques in ICA in CT images.

Our approach builds upon the mean shift paradigm via an adaptive thresholding strategy. The data consists of single CT slices from 75 patients, with variety of plaque sizes and number of calcium regions. The manual measurements were carried out by a neuroradiologist for benchmarking. The calcium burden was measured as the area of the labelled plaque. Various metrics were employed to compare manual and automated measurements including correlation coefficient (CC), dice similarity (DS), Jacard Index (JI), polyline distance metric (PDM) and precision of merit (PoM).

We found that our automated method of calcium area characterization performed accurately compared to manual measurements with CC=0.978, and PoM=0.915. The PDM, DS, and JI, also indicate a good performance with a mean DS=0.85 (SD=0.085), a mean JI=0.747 (SD=0.12), and a mean PDM=0.195 (SD=0.177).

The proposed approach for calcium burden measurement, yields reasonably accurate labelling of calcified plaque when benchmarked against manual measurements. The approach is independent of the number and size of calcium regions, and the prototype design shows encouraging results to be adaptable to clinical practice.