Related Experiment Videos
MSFCN-multiple supervised fully convolutional networks for the osteosarcoma segmentation of CT images
Lin Huang1, Wei Xia2, Bo Zhang3
1Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences, Suzhou, China; University of Science and Technology of China, Hefei, China.
Computer Methods and Programs in Biomedicine
|April 11, 2017
Summary
This study introduces a novel Multiple Supervised Fully Convolutional Network (MSFCN) for automatic osteosarcoma tumor segmentation on CT scans. The MSFCN method achieved superior accuracy in delineating tumor boundaries, outperforming other deep learning algorithms.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Osteosarcoma tumor segmentation on CT images is challenging due to significant variability in tumor size and structure.
- Accurate segmentation is crucial for effective treatment planning.
Purpose of the Study:
- To develop and evaluate an automatic tumor segmentation method for osteosarcoma using computed tomography (CT) images.
- To improve the accuracy and efficiency of tumor boundary delineation.
Main Methods:
- A novel Multiple Supervised Fully Convolutional Network (MSFCN) was developed for automatic segmentation.
- Image normalization was used as a pre-processing step.
- The MSFCN incorporated multiple supervised side output layers to guide multi-scale feature learning and capture both local and global image features.
Main Results:
- The MSFCN achieved an average Dice Similarity Coefficient (DSC) of 87.80%, sensitivity of 86.88%, Hammoude distance (HM) of 19.81, and F1-measure of 0.908.
- The MSFCN demonstrated superior performance compared to Fully Convolutional Networks (FCN), U-Net, and Holistically-Nested Edge Detection (HED) methods.
- The algorithm accurately segmented tumors with low contrast around soft tissue.
Conclusions:
- The proposed MSFCN algorithm enables fast and accurate delineation of osteosarcoma tumor boundaries on CT images.
- This technology has the potential to assist clinicians in developing more precise treatment strategies.
- The MSFCN method offers a significant advancement in automated medical image analysis for cancer diagnosis.