What type of study is this?

September 10, 2025Open Access

Lung Segmentation in X-ray Images of Tuberculosis Patients Using U-Net with CLAHE Preprocessing

IMIbnu Farid MabinaUniversitas Dian Nuswantoro CSChristy Atika SariUniversitas Dian Nuswantoro EREko Hari RachmawantoUniversitas Dian Nuswantoro

Key Points

U-Net model achieved pixel accuracy of 97.96% in segmenting lung regions from X-ray images.
Using CLAHE preprocessing significantly improved the quality of X-ray images affected by low contrast and noise.
The research utilized the Shenzhen and Montgomery datasets to validate the effectiveness of the U-Net model.
This approach may enhance early tuberculosis detection, reducing organ damage and disease spread.

Abstract

Tuberculosis (TB) is an infectious disease that commonly affects the lungs and remains one of the leading causes of death from infectious diseases. Early detection is essential to prevent further spread and organ damage. Chest X-ray images are one of the main methods for diagnosing TB, but image quality is often affected by low contrast and noise. This study proposes the application of Contrast Limited Adaptive Histogram Equalization (CLAHE) method to improve X-ray image quality, combined with U-Net deep learning architecture for lung segmentation in X-ray images of tuberculosis patients. U-Net was chosen due to its excellent capability in medical image segmentation, thanks to its architectural structure that has encoder-decoder with skip connections, which allows the model to retain detailed information on high-resolution images, even on complex and noisy data. Experimental results using the Shenzhen and Montgomery datasets show that the U-Net model with CLAHE achieves Pixel Accuracy 97.96%, Recall 94.93%, Specificity 98.97%, Dice Coefficient 95.87%, and Jaccard Index (IoU) 92.07%.

Ask AI

Helpful

Bookmark

View Full Paper