A U-Net Framework Using Differential Equations for Enhanced Computer Vision in Lung Disease Diagnosis

Articles in Press

Document Type : Research Paper

Authors

1 Department of Management Information Systems (MIS)‎, ‎School of Business‎, ‎King Faisal University (KFU)‎, ‎Al-Ahsa 31982‎, ‎Saudi Arabia.

2 Lankaran State University‎, ‎Lankaran‎, ‎Azerbaijan.

3 Department of System Programming‎, ‎South Ural State University‎, ‎Chelyabinsk‎, ‎454080‎, ‎Russia.

4 Department of Electric Drive‎, ‎Mechatronics and Electromechanics‎, ‎South Ural State University‎, ‎Chelyabinsk‎, ‎454080‎, ‎Russia.

5 College of Administration and Economics, Al-Iraqia University, Baghdad, Iraq.

Abstract

This study presents a U-Net-based approach for the classification of lung diseases using chest X-ray images. The model effectively leverages its encoder-decoder architecture and skip connections to capture both high-level semantic features and detailed spatial information, crucial for medical image analysis. The U-Net model was trained and tested on a dataset of 3,475 X-ray images, representing three classes: Normal, Lung Opacity, and Viral Pneumonia. The model achieved strong performance, with a weighted F1 score of 0.9770 and Cohen's Kappa of 0.9653, demonstrating its high accuracy in classifying lung diseases. These results confirm the suitability of U-Net for medical imaging tasks, particularly in detecting subtle abnormalities in chest X-ray images. However, the study also identifies challenges, including class imbalance in medical datasets and the computational demands of training large models like U-Net. Future improvements could focus on enhancing generalizability and reducing computational complexity through advanced data augmentation, domain adaptation, and architectural optimizations. Overall, this research highlights the potential of U-Net for developing reliable and efficient automated diagnostic tools in healthcare.

Keywords

Main Subjects

Computational Methods for Differential Equations

Articles in Press, Accepted Manuscript
Available Online from 26 November 2024

Files

History

  • Receive Date: 31 October 2024
  • Revise Date: 26 November 2024
  • Accept Date: 14 November 2024

Share

How to cite

Statistics