Abdominal multi-organ segmentation with cascaded convolutional and adversarial deep networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate abdominal organ segmentation is vital for medical applications like diagnosis and surgery.
• Current methods often require manual input or struggle with data scarcity.

Purpose of the Study:

• To develop a fully-automated deep learning model for multi-organ segmentation in abdominal CT and MR images.
• To improve the accuracy and generalization capability of abdominal organ segmentation.

Main Methods:

• Utilized a deep learning model extending conditional generative adversarial networks (cGANs).
• Employed cascaded, partially pre-trained convolutional encoder-decoders as the generator.
• Incorporated auto-context for simultaneous multi-level segmentation refinements.
• Fine-tuned encoders on non-medical images to address data limitations.

Main Results:

• Achieved state-of-the-art performance, outperforming existing encoder-decoder schemes.
• Secured first rank in three categories (liver CT, liver MR, multi-organ MR) of the CHAOS challenge.
• Demonstrated strong generalization capability for segmenting healthy liver, kidneys, and spleen.

Conclusions:

• The combination of cascaded convolutional and adversarial networks enhances automated multi-organ segmentation.
• The proposed pipeline offers improved guidance for clinicians in abdominal image interpretation.
• This deep learning approach shows significant potential for enhancing clinical decision-making.