Efficient Multiple Organ Localization in CT Image using 3D Region Proposal Network

Area of Science:

• Medical Image Analysis
• Computer Vision
• Radiology

Background:

• Organ localization is crucial for medical image analysis tasks like segmentation and registration.
• Existing methods often process CT images slice-by-slice, missing 3D spatial context.

Purpose of the Study:

• To develop an efficient and accurate 3D method for multiple organ localization in CT images.
• To improve upon existing convolutional neural network-based approaches.

Main Methods:

• A 3D region proposal network was utilized for direct 3D organ localization.
• A novel backbone network architecture was proposed to generate high-resolution feature maps.
• The method was evaluated on two clinical datasets covering 11 body and 12 head organs.

Main Results:

• The proposed 3D method achieved higher detection precision and localization accuracy compared to state-of-the-art methods.
• Processing speed was significantly improved, being 4 to 18 times faster.
• The method effectively leverages 3D spatial context for improved performance, especially for small organs.

Conclusions:

• The developed 3D region proposal network offers an efficient and accurate solution for multiple organ localization in CT images.
• This method enhances medical image analysis by providing a robust preprocessing step.
• Publicly available dataset and implementation are provided to facilitate further research.