Related Experiment Video
Updated: Dec 17, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Superpixel-based deep convolutional neural networks and active contour model for automatic prostate segmentation on
Giovanni L F da Silva1, Petterson S Diniz2, Jonnison L Ferreira2
1Applied Computing Group - NCA, Federal University of Maranhão - UFMA, Av. dos Portugueses, SN, Bacanga, São Luís, MA, 65085-580, Brazil. giovannilucca@nca.ufma.br.
Medical & Biological Engineering & Computing
|June 23, 2020
Summary
This study introduces a novel coarse-to-fine method for automatic prostate segmentation in 3D MRI scans. The approach enhances diagnostic accuracy by improving segmentation of the prostate gland.
Area of Science:
- Medical imaging
- Computer-aided diagnosis
- Biomedical engineering
Background:
- Accurate prostate segmentation is crucial for cancer diagnosis and treatment planning.
- Current methods face challenges due to unclear boundaries, variations in patient anatomy, and image resolution.
- Existing techniques like atlas-based, active contour, and deep learning models have limitations in accuracy and automation.
Purpose of the Study:
- To develop an automatic and robust coarse-to-fine segmentation method for prostate 3D MRI scans.
- To overcome limitations of existing prostate segmentation techniques.
- To improve the precision and reliability of automated prostate gland delineation.
Main Methods:
- A novel coarse-to-fine segmentation approach was developed for 3D prostate MRI.
- The coarse stage integrates local texture/spatial info using Intrinsic Manifold Simple Linear Iterative Clustering and a probabilistic atlas within a deep convolutional neural network, optimized by particle swarm optimization.
- The fine stage employs the 3D Chan-Vese active contour model for precise prostate surface extraction.
Main Results:
- The proposed method achieved a Dice Similarity Coefficient of 84.86% on the Prostate 3T and PROMISE12 databases.
- Performance metrics included a relative volume difference of 14.53%, sensitivity of 90.73%, specificity of 99.46%, and accuracy of 99.11%.
- The results demonstrate significant performance improvements over previously published methods.
Conclusions:
- The developed coarse-to-fine segmentation method shows high performance and potential for clinical application.
- This automated approach offers a reliable solution for prostate segmentation in 3D MRI.
- The findings suggest a promising advancement in computer-aided diagnosis for prostate cancer management.
