An efficient deep equilibrium model for medical image segmentation.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Accurate medical image segmentation is crucial for diagnosis and treatment planning.
• Existing methods, including classical algorithms and deep learning, have limitations in accuracy and generalizability.
• Nuclei segmentation in histopathological images presents a significant challenge due to variations in cell morphology and tissue types.

Purpose of the Study:

• To propose an effective method for medical image segmentation by combining classical approaches and deep learning.
• To enhance the accuracy, generalizability, and stability of segmentation models.
• To demonstrate the method's efficacy using nuclei segmentation in histopathological images.

Main Methods:

• Modeling neural networks as fixed-point iterations with a feedback loop to achieve system equilibrium.
• Formulating nuclei segmentation as a dynamic process searching for equilibrium.
• Dynamically driving segmentation towards expected values by iteratively refining outputs based on original images.

Main Results:

• Successfully segmented nuclei from cells in diverse histopathological images.
• Significantly increased model accuracy, generalizability, and stability compared to baseline methods.
• Demonstrated robust performance across different tissue types and open datasets.

Conclusions:

• The proposed method effectively integrates classical and deep learning techniques for medical image segmentation.
• The dynamic, feedback-driven approach enhances segmentation performance with minimal changes to the deep neural network backbone.
• This method offers a promising solution for challenging segmentation tasks in computational pathology.