A Convolutional Neural Network Architecture for Segmentation of Lung Diseases Using Chest X-ray Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate lung segmentation is crucial for diagnosing and treating lung diseases.
• Manual segmentation by radiologists is time-consuming and prone to variability.
• Deep learning offers automated solutions for medical image analysis.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for automated lung segmentation in chest X-ray images.
• To improve the efficiency and accuracy of lung segmentation compared to manual methods.

Main Methods:

• A convolutional neural network (CNN) architecture incorporating a concatenate block with a transpose layer was proposed.
• The model was trained and validated using k-fold cross-validation (k=5) on a Kaggle chest X-ray dataset.
• Hyperparameters including batch size (32), Adam optimizer, and 40 epochs were optimized.

Main Results:

• The proposed model achieved high performance metrics for lung segmentation.
• Key performance indicators included an accuracy of 0.97, Intersection over Union (IoU) of 0.93, and a Dice coefficient of 0.96.

Conclusions:

• The developed deep learning model demonstrates significant potential for accurate and automated lung segmentation from chest X-rays.
• This approach can assist radiologists in faster and more precise diagnosis of lung conditions.