Related Experiment Video
Updated: Jun 8, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Markov random field driven region-based active contour model (MaRACel): application to medical image segmentation
Jun Xu1, James P Monaco, Anant Madabhushi
1Department of Biomedical Engineering, Rutgers University, USA. junxu@rci.rutgers.edu
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|October 1, 2010
Summary
This study introduces a new Markov random field (MRF) driven active contour model (MaRACel) for improved medical image segmentation, outperforming existing models in accuracy.
Area of Science:
- Medical Image Analysis
- Computer Vision
- Computational Pathology
Background:
- Current region-based active contour (RAC) models for medical image segmentation often overlook spatial dependencies.
- Ignoring contextual information can limit the accuracy of segmentation in complex medical images.
Purpose of the Study:
- To develop a novel active contour model incorporating Markov random fields (MRFs) to leverage contextual information.
- To enhance medical image segmentation by addressing the statistical independence assumption in traditional RAC models.
Main Methods:
- Introduction of a Markov random field (MRF) driven region-based active contour model (MaRACel).
- Development of a continuous analogue to the discrete Potts model for integration into the variational framework.
- Comparison of MaRACel against Chan & Vese (CV) and Rousson & Deriche (RD) models using breast DCE-MR and prostate histopathology images.
Main Results:
- MaRACel demonstrated superior performance in segmenting cancerous lesions in breast DCE-MR images.
- In segmenting prostatic acini across 200 histopathology images, MaRACel achieved 71% sensitivity, 95% specificity, and 74% positive predictive value.
- CV and RD models showed significantly lower performance metrics (e.g., CV: 19% sensitivity, 81% specificity; RD: 53% sensitivity, 88% specificity).
Conclusions:
- The proposed MaRACel model effectively utilizes contextual information through MRF priors for more accurate medical image segmentation.
- MaRACel represents a significant advancement over existing CV and RD models, particularly for segmenting challenging structures like prostatic glands.
