Related Experiment Video
Updated: Jun 27, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Self and Mixed Supervision to Improve Training Labels for Multi-Class Medical Image Segmentation
Jianfei Liu1, Christopher Parnell2, Ronald M Summers1
1Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, Radiology and Imaging Sciences, National Institutes of Health Clinical Center, Bethesda, MD USA 20892.
Arxiv
|May 7, 2024
Summary
This study introduces an improved method for medical image segmentation by enhancing training labels using a dual-branch network and transfer learning. The approach significantly boosts segmentation accuracy for multi-class tasks, particularly in abdominal CT scans.
Area of Science:
- Medical image analysis
- Computer-aided diagnosis
- Artificial intelligence in medicine
Background:
- Accurate training labels are crucial for multi-class medical image segmentation but are expensive and time-consuming to create.
- Previous work developed a dual-branch network for single-class segmentation using weak and strong labels with self-supervision.
- Existing methods face challenges in generating high-quality labels for complex, multi-class segmentation tasks.
Conclusions:
- The combination of a dual-branch network and transfer learning offers an efficient strategy for enhancing training labels in multi-class medical image segmentation.
- This approach effectively addresses the challenges of costly and time-consuming manual annotation in medical imaging.
- The validated improvements suggest a promising direction for advancing automated medical image analysis and diagnostic tools.
