Enhancing semantic segmentation in chest X-ray images through image preprocessing: ps-KDE for pixel-wise substitution...

Area of Science:

• Medical Imaging
• Deep Learning
• Image Preprocessing

Background:

• Deep learning semantic segmentation in medical imaging aids disease classification.
• Contrast Limited Adaptive Histogram Equalization (CLAHE) improves segmentation but has limitations.
• Improved contrast enhancement is needed for diverse medical imaging datasets.

Purpose of the Study:

• To introduce and evaluate ps-KDE, a novel preprocessing technique for medical image contrast enhancement.
• To assess the impact of ps-KDE on deep learning-based organ segmentation in chest X-rays.
• To compare ps-KDE's performance against CLAHE in segmenting major thoracic organs.

Main Methods:

• A novel ps-KDE technique was developed, augmenting image contrast using normalized pixel frequencies.
• A U-Net architecture with a ResNet34 backbone was utilized for segmentation.
• Five models were trained to segment the heart, left lung, right lung, left clavicle, and right clavicle.

Main Results:

• The ps-KDE model achieved a significantly higher Dice score (0.780) for left lung segmentation compared to CLAHE (0.717), p<0.01.
• Ps-KDE demonstrated greater robustness, with CLAHE models showing misclassifications in some test cases.
• The ps-KDE algorithm is publicly available for use.

Conclusions:

• Ps-KDE offers superior performance over existing techniques like CLAHE for specific lung region segmentation.
• This improved segmentation can benefit downstream applications such as disease classification and risk stratification.
• The findings suggest ps-KDE is a valuable tool for enhancing medical image analysis.