Abdominal CT Segmentation for Body Composition Assessment Using Network Consistency Learning | JoVE Visualize

Area of Science:

Radiology and Medical Imaging
Artificial Intelligence in Medicine
Biomedical Image Analysis

Background:

Accurate skeletal muscle (SM) and adipose tissue estimation is crucial for prognostic assessment in various clinical scenarios.
Current body composition analysis often relies on computed tomography (CT) scans, requiring pixel-level semantic segmentation for precise SM estimation.
Estimating whole-body SM volume can be achieved by analyzing single 2D vertebral slices, but requires accurate segmentation.

Purpose of the Study:

To develop an efficient and cost-effective method for body composition assessment using limited labeled and abundant unlabeled CT images.
To improve the accuracy of semantic segmentation for skeletal muscle and adipose tissue estimation from abdominal CT scans.
To evaluate the effectiveness of Network Consistency Learning (NCL) in leveraging unlabeled data for enhanced segmentation performance.

Main Methods:

Trained a semantic segmentation model using both labeled and unlabeled abdominal CT slices.
Employed two identical segmentation networks with distinct weight initializations.
Implemented Network Consistency Learning (NCL) to enforce consistent predictions between the two networks, enabling learning from unlabeled data.

Main Results:

The proposed NCL approach achieved a 10% higher Dice Similarity Coefficient (DSC) compared to standard supervised segmentation networks.
Demonstrated the effectiveness of NCL in exploiting large amounts of unlabeled CT images for improved segmentation accuracy.
Successfully segmented abdominal CT images from a newly created in-house dataset.

Conclusions:

Network Consistency Learning (NCL) offers a significant improvement in body composition analysis from CT scans by effectively utilizing unlabeled data.
The proposed method provides an efficient and cost-effective solution for fast diagnosis, prognosis, and interventions.
This approach facilitates improved patient management through accurate body composition assessment.