A pathologist outlined the contour classifications from six endarterectomy patients. These sections were centered on the bifurcation with an average of 8-9 slices per patient. Fifty three sections out of these with matched MR image slices were chosen for analysis. The classification by the algorithm was then compared to the ground truth by histology. Each cap status per slice compared point by point for classification accuracy was used to calculate Pearson's correlation coefficients. The algorithm performs well in classifying thick and thin caps with

Thick cap

Thick cap

True

Figure 8.28: Correlation between true and classified pixels shows an R = 0.6442 for the thick cap (p value < 0.0001).

True

Figure 8.28: Correlation between true and classified pixels shows an R = 0.6442 for the thick cap (p value < 0.0001).

a correlation coefficient of 0.64 (significant with p value < 0.0001) and 0.62 (significant with p value < 0.0001) as shown in Figs. 8.28 and 8.29, respectively. The correlation coefficient for the ruptured cap is lesser (0.34, p value of 0.014) due to more false negatives and false positives. The correlation might be improved if specimen shrinkage [101] can be accounted for in matching correspondence between true and classified points. Differential shrinkage of the endarterectomy specimen during histological processing can cause twisting of the specimen around the arterial axis thus increasing classification error.

Thin cap

True

Figure 8.29: Correlation between true and classified pixels shows an R = 0.6245 for the thin cap (p value < 0.0001).

True

Figure 8.29: Correlation between true and classified pixels shows an R = 0.6245 for the thin cap (p value < 0.0001).

Was this article helpful?

## Post a comment