Pneumothorax Segmentation Using Low-Cost Computing Method

Diagnostic pneumothorax using Chest X-rays (CXR) poses a significant challenge due to the difficulty in visualizing subtle air accumulation and its potential overlap with other structures. While modern deep learning techniques hold promise for accurate pneumothorax segmentation, they often require substantial amounts of data and computational resources, which can be prohibitive in practical settings. Consequently, exploring traditional image processing methods remains a valuable approach to address this diagnostic challenge. This article purposes the efficacy of traditional image processing techniques, including Uniform Local Binary Patterns (ULBP), filtering, clustering, and manual pattern selection, for pneumothorax segmentation in CXR images. Our results demonstrate that these methods yield significantly lower Dice coefficient scores (below 0.20) in pneumothorax segmentation. Despite their limited performance, these methods hold potential for future development. To enhance their capabilities, further research is necessary to explore more robust feature extraction techniques, optimize parameter tuning, and incorporate advanced image processing algorithms.