Related Experiment Video
Updated: Dec 6, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
675
Y-Net for Chest X-Ray Preprocessing: Simultaneous Classification of Geometry and Segmentation of Annotations
Summary
This study presents a novel pre-processing method for chest X-rays using a modified Y-Net architecture. This approach normalizes image geometry and masks annotations, improving machine learning model robustness for medical diagnosis.
Area of Science:
- Medical Imaging
- Machine Learning
- Computer Vision
Background:
- Convolutional neural networks (CNNs) are leading algorithms for image classification and segmentation.
- Large medical imaging databases accelerate CNN use in biomedicine.
- Chest radiograph diagnosis relies on feature location, unlike general photograph classification.
Purpose of the Study:
- Introduce a general pre-processing step for chest X-ray input into machine learning algorithms.
- Develop a method to simultaneously learn chest geometry and segment radiographic annotations.
- Enhance the robustness of diagnostic algorithms by addressing reliance on annotations.
Main Methods:
- Utilized a modified Y-Net architecture with a VGG11 encoder.
- Trained the algorithm on 1000 manually labeled chest X-rays with augmentation.
- Developed a pre-processing step for normalizing geometry and masking annotations.
Main Results:
- Achieved acceptable geometry in 95.8% and annotation masks in 96.2% of cases (n=500).
- Control images showed significantly lower performance (27.0% geometry, 34.9% annotations).
- Expert clinicians evaluated the results.
Conclusions:
- The proposed pre-processing step demonstrates effectiveness in normalizing chest radiograph geometry and segmenting annotations.
- This method is hypothesized to improve the robustness of future diagnostic machine learning algorithms.
- This work provides a universal pre-processing technique for chest radiographs prior to further analysis.
