Semiautomatic tumor segmentation with multimodal images in a conditional random field framework.
Yu-Chi Hu1, Michael Grossberg2, Gikas Mageras3
1Memorial Sloan Kettering Cancer Center, Department of Medical Physics, 1275 York Avenue, New York, New York 10065, United States; City College of New York, Department of Computer Science, 160 Convent Avenue, New York, New York 10031, United States.
Journal of Medical Imaging (Bellingham, Wash.)
|July 15, 2016
Summary
This study introduces a semiautomatic algorithm for segmenting volumetric medical images. It combines multiple imaging modalities and human guidance for improved tumor segmentation accuracy.
Area of Science:
- Medical image analysis
- Computational imaging
- Radiology
Background:
- Volumetric medical imaging utilizes modalities like CT, MRI, and PET.
- Accurate segmentation of medical images is crucial for diagnosis and treatment planning.
- Existing methods often lack the integration of multimodal data and human oversight.
Purpose of the Study:
- To develop a semiautomatic segmentation algorithm for volumetric medical images.
- To leverage synergies between different imaging modalities and incorporate interactive human guidance.
- To improve the accuracy and efficiency of medical image segmentation, particularly for tumor detection.
Main Methods:
- A statistical segmentation framework using conditional random fields (CRF).
- Interactive training of statistical models with user-provided brush strokes for tumor/nontumor identification.
- Progressive segmentation achieved through graph-cut-based minimization, integrating multimodal data and prior segmentations.
Main Results:
- The algorithm demonstrates superior performance compared to fully automatic CRF and semiautomatic grow-cut methods.
- Validated using multimodal brain tumor segmentation challenge (BRATS 2012 and 2013) MRI datasets.
- The proposed method effectively integrates multimodal information and human interaction for enhanced segmentation.
Conclusions:
- The developed semiautomatic algorithm offers a novel approach to medical image segmentation.
- It successfully combines the strengths of different imaging modalities and human expertise.
- This method shows significant potential for improving clinical applications requiring precise image segmentation.
