Deep learning-aided 3D proxy-bridged region-growing framework for multi-organ segmentation

Zhihong Chen ^1,2, Lisha Yao ^2,3, Yue Liu ^1,4

⁰Institute of Computing Science and Technology, Guangzhou University, Guangzhou, 510006, China.

Scientific Reports +

|

April 29, 2024

View abstract on PubMed

Summary

This study introduces a 3D proxy-bridged region-growing framework for liver and spleen segmentation in CT images. The method achieves high accuracy with reduced annotation needs and lower GPU resource demands compared to deep learning approaches.

Area Of Science

Medical Imaging
Computer-Aided Diagnosis
Radiotherapy Planning

Background

Accurate 3D multi-organ segmentation in CT images is crucial for medical applications.
Current deep learning methods require extensive manual annotations and high GPU resources.
Challenges exist in efficient and resource-light 3D segmentation.

Purpose Of The Study

To develop a novel 3D framework for liver and spleen segmentation.
To reduce the reliance on manual annotations and high computational costs.
To improve the efficiency of computer-aided diagnosis and radiotherapy planning.

Main Methods

A 3D proxy-bridged region-growing framework is proposed.
Key slices are identified using intensity histograms for seed point calculation.
Segmentation is performed on superpixel images to mitigate noise, extended iteratively across slices.

Main Results

The framework achieved an average Dice Similarity Coefficient of ~0.93 for liver and spleen segmentation.
A Jaccard Similarity Coefficient of ~0.88 was obtained.
The method demonstrated comparable performance to deep learning models with reduced annotation and GPU requirements.

Conclusions

The proposed framework offers an efficient alternative for 3D liver and spleen segmentation.
It significantly lowers the demand for manual annotations and GPU resources.
This approach holds promise for enhancing computer-aided diagnosis and radiotherapy planning.

Keywords:

3D CT image Deep learning Multi-organ segmentation Proxy-bridging Region-growing