Gray-Level Co-Occurrence Matrix Texture Analysis of Breast Tumor Images in Prognosis of Distant Metastasis Risk
Gray-Level Co-Occurrence Matrix Texture Analysis of Breast Tumor Images in Prognosis of Distant Metastasis Risk
复制标题
DOI：
10.1017/s1431927615000379
复制
发表时间：
2015-06-01
影响因子：
2.8
通讯作者：
Radulovic, Marko
中科院分区：
文献类型：
--
作者：
Vujasinovic, Tijana;Pribic, Jelena;Radulovic, Marko
由于浸润性乳腺癌预后的特殊异质性，开发高度准确的预后工具以进行有效的治疗管理至关重要。基于这一迫切需求，我们旨在通过探索肿瘤组织学图像分析的预后价值来改善乳腺癌的预后。患者组（n = 78）的选择基于浸润性乳腺癌诊断，未接受全身治疗，中位随访时间为147个月。对原发性肿瘤组织切片数字图像进行回顾性灰度共生矩阵纹理分析，这些图像用苏木精和伊红非特异性染色或用上皮恶性细胞的泛细胞角蛋白抗体混合物特异性染色。单因素分析显示纹理分析与临床病理参数相比与转移风险有更强的相关性。将个体临床病理学和纹理变量组合成综合评分导致预后性能的进一步强大增强，准确度高达90%，曲线下面积[95%置信区间（CI）]的区分效率为0.94（0.87-0.99），风险比（95%CI）为20.1（7.5-109.4）。内部验证成功地进行了自助和分裂样本交叉验证，这表明该模型是可推广的。尽管需要在外部患者集上进行进一步验证，但这项初步研究表明，原发性乳腺肿瘤组织学纹理作为一种高度准确、简单且具有成本效益的远处转移风险预后指标具有潜在用途。
Owing to exceptional heterogeneity in the outcome of invasive breast cancer it is essential to develop highly accurate prognostic tools for effective therapeutic management. Based on this pressing need, we aimed to improve breast cancer prognosis by exploring the prognostic value of tumor histology image analysis. Patient group (n = 78) selection was based on invasive breast cancer diagnosis without systemic treatment with a median follow-up of 147 months. Gray-level co-occurrence matrix texture analysis was performed retrospectively on primary tumor tissue section digital images stained either nonspecifically with hematoxylin and eosin or specifically with a pan-cytokeratin antibody cocktail for epithelial malignant cells. Univariate analysis revealed stronger association with metastasis risk by texture analysis when compared with clinicopathological parameters. The combination of individual clinicopathological and texture variables into composite scores resulted in further powerful enhancement of prognostic performance, with an accuracy of up to 90%, discrimination efficiency by the area under the curve [95% confidence interval (CI)] of 0.94 (0.87-0.99) and hazard ratio (95% CI) of 20.1 (7.5-109.4). Internal validation was successfully performed by bootstrap and split-sample cross-validation, suggesting that the models are generalizable. Whereas further validation is needed on an external set of patients, this preliminary study indicates the potential use of primary breast tumor histology texture as a highly accurate, simple, and cost-effective prognostic indicator of distant metastasis risk.