Related Experiment Video
Updated: Nov 1, 2025
06:39
Three-dimensional Location Approach with Silk Thread Guided Laparoscopic Segmentectomy for Liver Tumor
Published on: May 23, 2025
204
DeepRecS: From RECIST Diameters to Precise Liver Tumor Segmentation
IEEE Journal of Biomedical and Health Informatics
|June 23, 2021
Summary
This study introduces an interactive deep learning method for liver tumor segmentation (LiTS) using Response Evaluation Criteria in Solid Tumor (RECIST) marks. The approach improves accuracy for hepatocellular carcinoma diagnosis and treatment planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Automated liver tumor segmentation (LiTS) methods struggle with the complex shapes and locations of hepatocellular carcinoma tumors.
- Clinical practice often relies on radiologists' manual estimations, such as Response Evaluation Criteria in Solid Tumors (RECIST) marks, to define tumor boundaries.
- Existing automated LiTS techniques lack the flexibility required for satisfactory clinical application.
Purpose of the Study:
- To develop an interactive deep learning (DL) based LiTS method guided by user-provided RECIST marks.
- To improve the accuracy and clinical utility of automated liver tumor segmentation.
- To address the limitations of current automated methods in modeling diverse tumor characteristics.
Main Methods:
- A three-step framework incorporating a RECIST mark propagation network (RMP-Net) to infer RECIST-like marks in unlabeled slices.
- A context-guided boundary-sensitive network (CGBS-Net) to extract tumor context and boundary information from RECIST marks.
- Post-processing using 3D conditional random field (CRF) and morphology hole-filling for refining segmentation maps.
Main Results:
- The proposed method demonstrated promising segmentation results on two clinical contrast-enhanced abdomen CT datasets.
- The RMP-Net effectively estimated RECIST-like marks in off-RECIST slices, aiding segmentation.
- The CGBS-Net successfully distilled crucial contextual and boundary information, leading to accurate tumor map prediction.
Conclusions:
- The interactive DL-based LiTS method, guided by RECIST marks, offers a significant advancement over traditional automated approaches.
- The integration of RECIST mark propagation and context-guided segmentation effectively handles tumor variability.
- This approach shows superior performance compared to state-of-the-art interactive segmentation methods, offering clinical potential for hepatocellular carcinoma management.
