CALSFOAM-completed automated local statistics based first order absolute moment" for carotid wall recognition, segmentation and IMT measurement: validation and benchmarking on a 300 patient database.

AIM

In this work we present a novel methodology (called CALSFOAM) for the automated segmentation of ultrasound carotid images and intima-media thickness (IMT) measurement. CALSFOAM was developed in order to overcome limitations of a previously developed snake-based technique.

METHODS

CALSFOAM consists of two stages: Stage-I is an automatic recognition of the carotid artery system in an image frame and Stage-II is a combination of segmentation and IMT measurement sub-system. Stage-I is performed by using local statistics and by automatically tracing the profile of the distal adventitia. Stage-II takes the traced adventitia boundary and builds an ROI for distal wall segmentation that uses a first order absolute moment (FOAM) technique. CALSFOAM was benchmarked against our previous snake based technique and validated on a 300-image multi-institutional dataset.

RESULTS

CALSFOAM's lumen-intima (LI) segmentation error was 0.049±0.039 mm, the media-adventitia (MA) error was 0.088±0.054 mm; the IMT measurement bias was 0.125±0.103 mm. To reduce CALSFOAM error, we adopted a GREEDY approach for fusing the boundaries from the two techniques and obtained LI and MA errors equal to 0.02±0.014 mm, 0.023±0.013 mm, and an IMT bias of 0.074±0.068 mm.

CONCLUSION

Even though CALSFOAM's performance was lower than snake-based segmentation techniques, it helped in avoiding possible inaccuracies of snakes and its parameter sensitivities. The very accurate performance obtained by the GREEDY approach demonstrated that the two techniques could be considered as complementary.