TY - JOUR
T1 - An Accurate Segmentation Framework for Static Ultrasound Images of the Gestational Sac
AU - Yin, Chenghuan
AU - Wang, Yu
AU - Zhang, Qixin
AU - Han, Fangfang
AU - Yuan, Zhengwei
AU - Yao, Yudong
N1 - Publisher Copyright:
© 2021, Taiwanese Society of Biomedical Engineering.
PY - 2022/2
Y1 - 2022/2
N2 - Purpose: Abnormal morphology characteristics of the gestational sac give accurate prediction for the early spontaneous abortion. However, due to the high noise and the weak transmission of ultrasonic signals, accurate segmentation is an urgent and challenging task for quantitative morphology analysis of gestational sacs in ultrasound images. Methods: This paper proposes an accurate segmentation framework for gestational sac ultrasound images from two different scanning machines based on an improved level-set algorithm driven by shape domain-specific knowledge. Firstly, the gestational sac candidate image is roughly segmented using the Chan-Vese (CV) model, and the minimum convex polygon is reserved according to the quasi-circular character of the sacs. Then, the gestational sacs are divided into two categories, such as the concave and convex ones, followed by separate corresponding processing for further accurate segmentation. Results: A total of 194 ultrasound images of the gestational sacs of 6–9 were processed in this experiment. For testing, we have obtained the mean Dice and Intersection over Union (IOU) value of 0.916 and 0.842, and the average sensitivity (SEN) and positive prediction rate (PPV) are 0.958 and 0.890, respectively. The evaluation results show that our proposed segmentation framework has better performance than other commonly used segmentation methods. Conclusion: The evaluation results show that our proposed segmentation framework has better performance than other commonly used segmentation methods.
AB - Purpose: Abnormal morphology characteristics of the gestational sac give accurate prediction for the early spontaneous abortion. However, due to the high noise and the weak transmission of ultrasonic signals, accurate segmentation is an urgent and challenging task for quantitative morphology analysis of gestational sacs in ultrasound images. Methods: This paper proposes an accurate segmentation framework for gestational sac ultrasound images from two different scanning machines based on an improved level-set algorithm driven by shape domain-specific knowledge. Firstly, the gestational sac candidate image is roughly segmented using the Chan-Vese (CV) model, and the minimum convex polygon is reserved according to the quasi-circular character of the sacs. Then, the gestational sacs are divided into two categories, such as the concave and convex ones, followed by separate corresponding processing for further accurate segmentation. Results: A total of 194 ultrasound images of the gestational sacs of 6–9 were processed in this experiment. For testing, we have obtained the mean Dice and Intersection over Union (IOU) value of 0.916 and 0.842, and the average sensitivity (SEN) and positive prediction rate (PPV) are 0.958 and 0.890, respectively. The evaluation results show that our proposed segmentation framework has better performance than other commonly used segmentation methods. Conclusion: The evaluation results show that our proposed segmentation framework has better performance than other commonly used segmentation methods.
KW - Accurate segmentation
KW - Chan-Vese model
KW - Concave and convex polygon
KW - Gestational sac
KW - Ultrasound image
UR - http://www.scopus.com/inward/record.url?scp=85123099296&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85123099296&partnerID=8YFLogxK
U2 - 10.1007/s40846-021-00674-4
DO - 10.1007/s40846-021-00674-4
M3 - Article
AN - SCOPUS:85123099296
SN - 1609-0985
VL - 42
SP - 49
EP - 62
JO - Journal of Medical and Biological Engineering
JF - Journal of Medical and Biological Engineering
IS - 1
ER -