Shape Reconstruction for Abdominal Organs based on a Graph Convolutional Network.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• High-resolution medical imaging (CT, MRI) is limited during surgery/radiotherapy, yielding low-resolution cone-beam CT and X-ray images.
• Accurate organ segmentation is challenging for air-filled structures like the duodenum and stomach, even with high-resolution CT.
• Detecting and reconstructing these organs is crucial for precise medical interventions.

Purpose of the Study:

• To propose a novel graph convolutional network (GCN) based method for reconstructing organs that are difficult to detect in medical images.
• To leverage features from surrounding detectable organs to infer the shape and location of target organs.
• To validate the method's performance using both single and multiple organ mesh experiments.

Main Methods:

• A graph convolutional network (GCN) approach was developed to reconstruct target organs.
• The method utilizes features from adjacent, detectable organs to predict the shape and position of the target organ.
• Mesh deformation parameters are learned and applied to an organ template, which provides an initial topological structure.

Main Results:

• The proposed GCN method demonstrated effectiveness in reconstructing organs that are challenging to segment.
• Experiments with single and multiple organ meshes confirmed the method's performance.
• The approach successfully utilized surrounding organ features to guide reconstruction.

Conclusions:

• The GCN-based method offers a promising solution for reconstructing difficult-to-detect organs in medical imaging.
• This technique can improve the accuracy of organ localization and shape estimation, particularly in low-resolution scenarios.
• The findings have implications for enhanced surgical and radiotherapy planning.