Silva Suane Pires Pinheiro da, Ivo Roberto Fernandes, Barroso Calleo Belo, Fernandes João Carlos Nepomuceno, Portela Thiago Ferreira, Medeiros Aldísio Gonçalves, Sousa Pedro Henrique F de, Song Houbing, Rebouças Filho Pedro Pedrosa
Department of Teleinformatics Engineering, Federal University of Ceará (UFC), Fortaleza 60440-900, CE, Brazil.
Federal Institute of Education, Science and Technology of Ceara (IFCE), Fortaleza 60040-531, CE, Brazil.
Sensors (Basel). 2025 Apr 30;25(9):2864. doi: 10.3390/s25092864.
Lung diseases, including chronic obstructive pulmonary disease (COPD) and pulmonary fibrosis, pose significant health challenges due to their high morbidity and mortality rates. Computed tomography (CT) scans play a critical role in early diagnosis and disease management, yet traditional segmentation methods often falter in addressing anatomical variability and pathological complexity. To overcome these limitations, this study introduces the context-driven active contour (CDAC), a new segmentation method that combines active contour models (ACMs) with contextual analysis. CDAC leverages contextual information from image embeddings and expert annotations to refine segmentation precision. The algorithm employs contextual attention force (CAF) as an external energy term and contextual balloon force (CBF) as an internal energy term, enabling robust contour adaptation. Evaluations were conducted on CT images of healthy lungs, as well as those affected by COPD and pulmonary fibrosis. CDAC achieved notable performance metrics, including a Dice coefficient of 96.8% for healthy lungs, an Accuracy of 94.5% for COPD, and a Jaccard Index of 92.3% for pulmonary fibrosis. These results demonstrate the method's effectiveness and adaptability. By integrating contextual insights, CDAC offers a promising solution for enhancing computer-aided diagnostic (CAD) systems in the management of lung diseases.
肺部疾病，包括慢性阻塞性肺疾病（COPD）和肺纤维化，因其高发病率和死亡率而对健康构成重大挑战。计算机断层扫描（CT）扫描在早期诊断和疾病管理中起着关键作用，但传统的分割方法在应对解剖学变异性和病理复杂性时往往表现不佳。为了克服这些局限性，本研究引入了上下文驱动主动轮廓（CDAC），这是一种将主动轮廓模型（ACM）与上下文分析相结合的新分割方法。CDAC利用来自图像嵌入和专家注释的上下文信息来提高分割精度。该算法采用上下文注意力力（CAF）作为外部能量项，上下文气球力（CBF）作为内部能量项，实现强大的轮廓自适应。对健康肺部以及受COPD和肺纤维化影响的肺部的CT图像进行了评估。CDAC取得了显著的性能指标，包括健康肺部的Dice系数为96.8%，COPD的准确率为94.5%，肺纤维化的Jaccard指数为92.3%。这些结果证明了该方法的有效性和适应性。通过整合上下文见解，CDAC为增强肺部疾病管理中的计算机辅助诊断（CAD）系统提供了一个有前景的解决方案。
