A Multiorgan Segmentation Model for CT Volumes via Full Convolution-Deconvolution Network.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate abdominal organ segmentation is crucial for medical diagnosis and treatment planning.
• Existing segmentation methods often struggle with capturing fine organ details and handling inter-subject variability in CT volumes.

Purpose of the Study:

• To develop and evaluate a novel two-stage deep learning model for improved abdominal organ segmentation on CT volumes.
• To enhance the accuracy of segmenting key abdominal organs like the liver, spleen, and kidney.

Main Methods:

• A two-stage approach combining a full convolution-deconvolution network (FCN-DecNet) for initial segmentation and a multiscale weights probabilistic atlas (MS-PA) for refinement.
• The FCN-DecNet incorporates specialized unpooling, deconvolutional, and fusion layers to capture organ details.
• The MS-PA optimizes coarse segmentation using spatial and intensity characteristics of atlases, leveraging inter-subject variability.

Main Results:

• The proposed model achieved high segmentation accuracy for abdominal organs.
• Specific Dice index scores were reported: 90.1 ± 1% for the liver, 89.0 ± 1.6% for the spleen, and 89.0 ± 1.3% for the kidney.
• The model effectively utilizes spatial location and gray information for accurate segmentation.

Conclusions:

• The developed coarse-fine model demonstrates superior performance in abdominal organ segmentation from CT volumes.
• This approach offers a robust solution for precise liver, spleen, and kidney segmentation, with potential clinical applications.