Method details

Download PDF file of article
Improving Accuracy in Coronary Lumen Segmentation via Explicit Calcium Exclusion, Learning-based Ray Detection and Surface Optimization



Abstract:
Since the diagnostic value of non-invasive procedures for diagnosing CAD relies on an accurate extraction of the lumen, a precise segmentation of the coronary arteries is crucial. As manual segmentation is tedious, time-consuming and subjective, automatic procedures are desirable. We present a novel fully-automatic method to accurately segment the lumen of coronary arteries in the presence of calcified and non-calcified plaque. Our segmentation framework is based on three main steps: boundary detection, calcium exclusion and surface optimization. A learning-based boundary detector enables a robust lumen contour detection via dense ray-casting. The exclusion of calcified plaque is assured through a novel calcium exclusion technique which allows us to accurately capture stenoses of diseased arteries. The boundary detection results are then incorporated into a closed set formulation whose minimization yields an optimized lumen surface. On the standardized Rotterdam evaluation framework, a segmentation accuracy is achieved which is comparable to clinical experts and superior to current automatic methods. Averaged over 30 datasets, we obtain a DICE overlap of 74 (73)%, an MSD of 0.35 (0.55) mm and MAXSD of 2.99 (3.73) mm on diseased (healthy) vessel segments as compared to three medical experts.

Detection confusion tables

Calc. cat.QCA (per segment)CTA (per lesion)
 TPFPFNTP+FPTP+FNTPFPFNTP+FPTP+FN
All  17  52  11  69  28  22  66  25  88  47 
The results of this method are based on the following centerlines: own or none.

Detection (QCA per segment / CTA per lesion)

Calc. cat.QCA
Sens.
QCA
P.P.V.
CTA
Sens.
CTA
P.P.V.
Avg. rank
 %rank%rank%rank%rank 
0 (0 - 0.1)  60.0  4.0  30.0  4.0  57.1  5.0  33.3  6.0  4.8 
1 (0.1 - 10)  100.0  1.0  25.0  5.0  40.0  11.0  28.6  10.0  6.8 
2 (11 - 100)  42.9  11.0  30.0  6.0  27.3  14.0  21.4  10.0  10.2 
3 (101 - 400)  80.0  3.0  25.8  8.0  57.9  4.0  29.7  9.0  6.0 
4 (400+)  40.0  12.0  14.3  17.0  40.0  11.0  11.1  15.0  13.8 
All  60.7  6.0  24.6  7.0  46.8  11.0  25.0  11.0  8.8 
For ranking  60.7  6.0  24.6  7.0  46.8  11.0  25.0  11.0  8.8 
These results are based on 30 datasets and 19 submissions.


Quantification

Calc. cat.QCA
Avg. Abs. diff.
QCA
R.M.S. diff.
CTA
Weigthed Kappa
Avg. rank
 %rank%rankΚrank 
0 (0 - 0.1)  42.6  10.0  48.7  10.0  0.28  6.0  8.0 
1 (0.1 - 10)  37.2  11.0  40.7  11.0  0.26  7.0  9.0 
2 (11 - 100)  46.4  10.0  53.7  11.0  0.30  7.0  8.8 
3 (101 - 400)  47.5  10.0  52.8  10.0  0.32  6.0  8.0 
4 (400+)  67.1  15.0  71.3  15.0  0.26  10.0  12.5 
All  49.0  11.0  55.1  12.0  0.30  7.0  9.2 
For ranking  49.0  11.0  55.1  12.0  0.30  7.0  9.2