Related Experiment Video
Updated: Feb 28, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Simultaneous Tumor Segmentation, Image Restoration, and Blur Kernel Estimation in PET Using Multiple Regularizations
Laquan Li1, Jian Wang1, Wei Lu2,3
1Key Laboratory of Image Processing and Intelligent Control of Ministry of Education of China, School of Automation, Huazhong University of Science and Technology, Wuhan 430074, China.
Computer Vision and Image Understanding : CVIU
|June 13, 2017
Summary
This study introduces a novel variational method to simultaneously restore PET images and segment tumors, improving accuracy by addressing partial volume effects. The method enhances tumor delineation for radiation oncology applications.
Area of Science:
- Medical Imaging
- Image Processing
- Radiation Oncology
Background:
- Partial volume effect (PVE) significantly degrades PET image quality and tumor segmentation accuracy.
- Accurate tumor segmentation is critical for radiation oncology applications.
- Image restoration and tumor segmentation are interdependent processes.
Purpose of the Study:
- To develop a variational method for simultaneous PET image restoration and tumor segmentation.
- To address the challenges posed by PVE in PET imaging.
- To improve the accuracy of tumor delineation in radiation oncology.
Main Methods:
- Integrated total variation (TV) semi-blind de-convolution with Mumford-Shah segmentation.
- Employed TV regularization on tumor edges and L2 regularization within tumor regions.
- Modeled the blur kernel as an anisotropic Gaussian.
- Utilized a Γ-convergence approximation and alternating minimization (AM) algorithm for optimization.
Main Results:
- Achieved high performance in simultaneous image restoration, tumor segmentation, and blur kernel estimation.
- Phantom study showed recovery coefficients (RC) close to 1, indicating effective image recovery.
- Clinical datasets yielded average Dice Similarity Indexes (DSIs) of 0.79 and 0.80 for tumor segmentation.
- Estimated blur kernel widths had relative errors <19% (transverse) and <7% (axial).
Conclusions:
- The proposed variational method effectively performs simultaneous PET image restoration and tumor segmentation.
- The method demonstrates robustness in handling PVE and accurately delineating tumors.
- Results support the clinical utility of the method in radiation oncology for improved treatment planning.