Enhancing U-Net Segmentation Accuracy Through Comprehensive Data Preprocessing.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of lung regions in CT scans is crucial for diagnosing lung diseases such as COPD and COVID-19.
• Automated analysis of lung diseases relies heavily on precise segmentation of lung structures in medical imaging.
• Existing segmentation methods may struggle with artifacts and variations in CT image quality.

Purpose of the Study:

• To enhance the accuracy of U-Net segmentation models for lung regions in CT scans.
• To develop and validate a robust preprocessing pipeline for medical image segmentation.
• To improve the reliability of automated lung disease analysis through optimized data preparation.

Main Methods:

• A preprocessing pipeline involving CT image normalization, binarization, and morphological operations was developed.
• Region-of-interest (ROI) filtering was applied to effectively isolate lung areas.
• The preprocessed data was used to train and evaluate U-Net segmentation models.

Main Results:

• The preprocessing pipeline significantly improved segmentation quality by providing clean, consistent input data.
• Intersection over Union (IoU) and Dice coefficients exceeded 0.95 on training datasets.
• Experimental results validated the effectiveness of the proposed preprocessing strategy.

Conclusions:

• Preprocessing is a critical standalone step for optimizing deep learning-based medical image analysis.
• The developed pipeline enhances the accuracy of lung segmentation in CT scans.
• This approach holds promise for improving automated diagnosis and analysis of lung diseases.