Fig. 1 | Scientific Reports

Fig. 1

From: Evaluating the generalizability of an automated coronary artery calcium segmentation and scoring algorithm using multi-vendor dataset

Fig. 1

Visualization of the per-lesion segmentation performance of RICAU-Nets on a variety of input images. The input images were sourced from eight patients, with two patients represented in each group. Ground truths and predictions from the test sets are presented for group 1 (a, b), in group 2 (c, d), in group 3 (e, f), and in group 4 (g, h). The per-lesion Dice scores for each image are as follows: LM: 99.70%, LAD: 97.96% for (a); LCX: 99.20% for (b); LM: 99.32%, LAD: 94.18%, LCX: 65.56% for (c); LAD: 100%, LCX: 95.24%, RCA: 100% for (d); LAD: 100%, RCA: 100% for (e); LAD: 81.63% for (f); LM: 98.73%, LAD: 100% for (g); LAD: 85.38% for (h). Common misclassifications in the segmentation task are observed in (c, e, f, h). The corresponding calcium lesions are displayed in the colored boxes, with their names highlighted in the top right corner of the images.

Back to article page