Learning With Context Feedback Loop for Robust Medical Image Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep learning, particularly convolutional neural networks (CNNs), is widely used for medical image segmentation.
• Traditional CNN approaches may produce incomplete or unrealistic segmentations due to limited pixel interdependence.

Purpose of the Study:

• To develop a fully automatic and robust deep learning method for medical image segmentation.
• To improve the accuracy and anatomical plausibility of segmentation results.

Main Methods:

• A recurrent framework utilizing two interconnected systems: an encoder-decoder CNN and a fully convolutional network (FCN)-based context feedback system.
• The FCN feedback loop integrates high-level features back into the CNN's learning process to correct errors and refine predictions.

Main Results:

• The proposed method achieved superior performance on four diverse clinical datasets, outperforming existing state-of-the-art techniques.
• Demonstrated robustness in segmenting single and multi-structures, even in low-contrast images.
• Results showed improved anatomical plausibility and accuracy over time with the feedback mechanism.

Conclusions:

• Formulating medical image segmentation as a recurrent framework with a context feedback loop offers a robust and efficient approach.
• This method enhances deep learning capabilities for producing anatomically plausible and reliable medical image segmentations.