A deep learning based dual encoder-decoder framework for anatomical structure segmentation in chest X-ray images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Automated segmentation of chest X-ray fluoroscopy is crucial for computer-aided diagnosis (CAD).
• Challenges include indistinct structures, anatomical variations, medical devices, and image artifacts.
• Existing methods struggle with precise segmentation of multiple organs.

Purpose of the Study:

• To develop a robust deep learning framework for anatomical structure segmentation in chest radiographs.
• To enhance the accuracy of multi-organ and single-organ segmentation.
• To address the limitations of current CAD systems in chest X-ray analysis.

Main Methods:

• Proposed a dual encoder-decoder convolutional neural network (CNN) framework.
• Utilized a pre-trained VGG19 encoder with Squeeze-and-Excitation (SE) for feature calibration.
• Incorporated recurrent residual blocks and an attention gate module in the second decoder for contextual feature capture.
• Evaluated on three public chest X-ray datasets for multi-organ (heart, lungs, clavicles) and single-organ (lungs) segmentation.

Main Results:

• The proposed deep learning framework demonstrated superior performance in multi-organ and single-organ segmentation tasks.
• Achieved higher accuracy compared to existing multi-class and single-class segmentation methods.
• Effectively handled challenges like indistinct structures and anatomical variations.

Conclusions:

• The developed dual encoder-decoder CNN framework offers a robust solution for anatomical structure segmentation in chest radiographs.
• This approach significantly improves the accuracy of computer-aided diagnosis systems.
• The method shows promise for advancing automated analysis in medical imaging.