Vertebrae Segmentation with Generative Adversarial Networks for Automatic Cobb Angle Measurement.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Spinal Deformities

Background:

• Scoliosis diagnosis relies on Cobb angle (CA) measurement, typically requiring manual vertebrae identification.
• Manual CA calculation is time-consuming and prone to inaccuracies due to image artifacts and variable image quality.
• Accurate CA measurement is critical for effective scoliosis intervention and treatment.

Purpose of the Study:

• To evaluate Generative Adversarial Networks (GANs) for vertebrae segmentation in spinal X-rays.
• To compare GAN performance against traditional segmentation models for improved boundary definition.
• To assess the potential of GANs in developing a more robust CA estimation pipeline.

Main Methods:

• Utilized Pix2Pix ensemble models, a type of GAN, for vertebrae segmentation on spinal X-ray images.
• Compared the performance of GAN-based segmentation with traditional segmentation approaches.
• Focused on the ability to distinguish boundaries between individual vertebrae.

Main Results:

• Pix2Pix ensemble models demonstrated superior performance over traditional models in segmenting vertebrae.
• The GAN approach provided more distinct and defined vertebrae boundaries.
• Improved segmentation accuracy is crucial for reliable Cobb angle calculation.

Conclusions:

• GANs, particularly Pix2Pix ensembles, show promise for enhancing vertebrae segmentation in scoliosis imaging.
• This advancement can lead to more accurate and automated Cobb angle measurements.
• The developed method offers potential for a more robust CA estimation pipeline adaptable to diverse image qualities.