孤立性肺结节CT影像的计算机视觉分析

摘要

Aim:To discuss the diagnostic value of the characteristics of computer vision for the solitary pulmonary nodules. Methods: 89 cases of solitary pulmonary nodules confirmed by pathology were selected along with the CT image data obtained under high resolution scan. All cases were processed with multivariant mophological filtering and double segmentation. These characteristics of computer vision were used to distinguish magnetic nodules from the benign nodules after Hough function transformation and skeleton extraction. Results: The data of MaxRho indicated that the MaxRho value of cancerous noduleswas (141. 79 ± 8. 332) and that of benign nodules was (83. 27 ± 18. 35 ) , The data has statistical significance ( t=2. 339 ,P =0. 02). The Max-valley-sum of magnetic malignant nodules and benign nodules was (232. 33 ± 26. 73) and (93. 60 ± 22. 34 ) respectively (t = 4.433, P = 0.005). The value of Eul of cancerous and benign nodules were (9.87 ±2.57) and (4.18 ± 1.32) respectively (t =4.523 ,P =0.003). Conclusion: The diagnostic accuracy of these index were evaluated by receiver operator characteristic curve ( ROC) analysis. The characteristics of computer vision, included MaxRho, Max-valley-sum, and eul number can be used to distinguish malignant nodules from benign nodules.%目的:探讨电子计算机视觉分析对孤立性肺结节的诊断价值.方法:89例经手术病理证实的肺部结节电子计算机断层扫描(CT)图像经过形态学滤波以及二次图像分割,进行标准哈夫(Hough)函数变换,以提取的计算机视觉特征数据来研究孤立性肺结节的影像诊断.结果:恶性结节的最大极径(MaxRho)为(141.79±8.332),良性结节的MaxRho为(83.27±18.35),两者有统计学差异(t=2.339,P=0.02).恶性结节的峰值为(232.33±26.73),良性结节的峰值为(93.60±22.34),两者有统计学的明显差异(t=4.433,P =0.005).肺恶性结节的欧拉数为(9.87±2.57),良性结节为(4.18±1.32),两者有统计学的明显差异(t=4.523,P=0.003).结论:经受试者工作特征(ROC)曲线分析,计算机视觉分析对孤立性肺结节的鉴别具有较高的诊断价值.