Segmentation of 2D fetal ultrasound images by exploiting context information using conditional random fields

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Fetal ultrasound (US) image segmentation is challenging due to high noise and low contrast.
• Similar textures and gray levels between fetal structures and surrounding tissues complicate segmentation.
• Accurate segmentation is crucial for prenatal diagnosis and monitoring.

Purpose of the Study:

• To propose a novel Conditional Random Field (CRF) based framework for accurate fetal ultrasound image segmentation.
• To leverage contextual information of maternal tissues, amniotic fluid, and placenta for improved segmentation.
• To address inherent challenges in US imaging, including noise and low contrast.

Main Methods:

• Development of a CRF-based segmentation framework tailored for fetal ultrasound images.
• Utilization of wavelet-based texture features for image representation.
• Integration of Support Vector Machines (SVM) for initial label prediction within the CRF model.

Main Results:

• The proposed CRF framework demonstrated promising initial results on real-world fetal ultrasound datasets.
• The method effectively handled image noise and the similarity between fetal structures and their surroundings.
• Quantitative and qualitative improvements in segmentation accuracy were observed.

Conclusions:

• The novel CRF framework offers a robust solution for segmenting challenging fetal ultrasound images.
• Exploiting contextual information significantly enhances segmentation accuracy in noisy US environments.
• This approach holds potential for improving clinical assessments in prenatal care.