![]() |
![]() |
Method details
FrenchCoast: Fast, Robust Extraction for the Nice CHallenge on COronary Artery Segmentation of the Tree
Abstract:
This paper describes the pipeline for a fully automatic analysis of the coronary arteries from CTA data for the coronary artery evaluation framework. It consists of three consecutive steps. First, a tree extraction and segment labeling step is performed. Second, the labeled
segments are extracted and segmented by a longitudinal contour detection on the straightened MPR images. These longitudinal contours initialize the transversal contour detection scheme. Third, lesions are defined on the quantified segments with reference markers initially set at the beginning and the end of the defined segments.
Training datasets were used to tune the automatic lesion detection and
quantification in the third step. After this, test datasets were segmented
and quantified to evaluated the pipeline. The results are presented in
three categories: lesion detection, lesion quantification, and lumen border
segmentation. An average ranking is calculated for each category that
enables comparison with the ground truth data.
For the detection stage, a sensitivity of 19% and a PPV of 15% is achieved
as compared to QCA, while a sensitivity of 28% and a PPV of 36% is
achieved as compared to CTA. Moreover, the stenoses are quantified with
an averaged absolute difference of 33.4% as compared to QCA. Finally, a
Dice of 66% and 68% is obtained for diseased and healthy vessel segments
respectively, which is comparable to the observer’s performance.
Detection confusion tables
Calc. cat. | QCA (per segment) | CTA (per lesion) | ||||||||
---|---|---|---|---|---|---|---|---|---|---|
  | TP | FP | FN | TP+FP | TP+FN | TP | FP | FN | TP+FP | TP+FN |
All | 7 | 30 | 21 | 37 | 28 | 13 | 29 | 34 | 42 | 47 |
The results of this method are based on the following centerlines: LKEB team auto.
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) | 20.0 | 11.0 | 20.0 | 8.0 | 28.6 | 11.0 | 40.0 | 4.0 | 8.5 |
1 (0.1 - 10) | 0.0 | 16.0 | 0.0 | 16.0 | 40.0 | 11.0 | 100.0 | 1.0 | 11.0 |
2 (11 - 100) | 0.0 | 19.0 | 0.0 | 19.0 | 18.2 | 16.0 | 20.0 | 11.0 | 16.2 |
3 (101 - 400) | 30.0 | 15.0 | 20.0 | 11.0 | 15.8 | 17.0 | 16.7 | 13.0 | 14.0 |
4 (400+) | 60.0 | 4.0 | 42.9 | 6.0 | 80.0 | 4.0 | 57.1 | 4.0 | 4.5 |
All | 25.0 | 16.0 | 18.9 | 12.0 | 27.7 | 15.0 | 31.0 | 7.0 | 12.5 |
For ranking | 25.0 | 16.0 | 18.9 | 12.0 | 27.7 | 15.0 | 31.0 | 7.0 | 12.5 |
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 | % | rank | Κ | rank |   |
0 (0 - 0.1) | 27.6 | 3.0 | 35.7 | 5.0 | 0.20 | 9.0 | 6.5 |
1 (0.1 - 10) | 34.3 | 8.0 | 37.8 | 9.0 | 0.24 | 9.0 | 8.8 |
2 (11 - 100) | 34.6 | 7.0 | 41.8 | 8.0 | 0.27 | 9.0 | 8.2 |
3 (101 - 400) | 33.5 | 7.0 | 39.7 | 7.0 | 0.20 | 10.0 | 8.5 |
4 (400+) | 33.2 | 7.0 | 39.9 | 6.0 | 0.51 | 5.0 | 5.8 |
All | 32.5 | 8.0 | 39.3 | 7.0 | 0.27 | 9.0 | 8.2 |
For ranking | 32.5 | 8.0 | 39.3 | 7.0 | 0.27 | 9.0 | 8.2 |