摘要
目的:通过Logistic回归分析乳腺非肿块型病变（NMLs）的超声特征，建立NMLs的恶性风险列线图预测模型，提高超声医师对NMLs的诊断能力。方法:回顾性分析西京医院488人共493例NMLs的超声图像特征，以7:3的比例分为训练集和内部验证集，独立的外部验证集72例来自浙江省台州医院。筛出NMLs的恶性危险因素，构建列线图预测模型，采用ROC曲线和校准曲线评价该模型。结果:多因素分析显示钙化、结构扭曲、年龄、病变大小是NMLs的恶性独立危险因素（P<0.05）。列线图预测模型的训练集、内部验证集、外部验证集AUC分别为0.91、0.89、0.94，敏感度分别为86%、86%、95%，特异度分别为80%、77%、87%。模型的校准曲线表现出良好的一致性，平均绝对误差分别为0.014、0.034和0.058。结论:基于上述独立危险因素构建的恶性风险列线图预测模型具有可靠的临床参考价值，可以协助超声医师提高对NMLs诊断能力。
Abstract
Objective:To analyze the ultrasound features of non-mass lesions (NMLs) of the breast by logistic regression, to establish a predictive model for the malignancy risk of NMLs in nomogram, and to improve the diagnostic ability of sonographers for NMLs.Methods:The ultrasound image features of 493 cases of NMLs from 488 people in Xijing Hospital were retrospectively analyzed and divided into a training set and an internal validation set in a ratio of 7:3, and an independent external validation set of 72 cases from Taizhou Hospital in Zhejiang Province. The malignant risk factors of NMLs were screened out, a nomogram prediction model was constructed, and the model was evaluated using ROC and calibration curves.Results:Multifactorial analysis showed that calcification, structural distortion, age, and size were independent risk factors for malignancy in NMLs (P<0.05). The training set, internal validation set, and external validation set AUCs of the nomogram prediction model were 0.91, 0.89, and 0.94, respectively, with sensitivities of 86%, 86%, and 95%, and specificities of 80%, 77%, and 87%. The calibration curves of the model showed good agreement, with mean absolute errors of 0.014, 0.034, and 0.058, respectively.Conclusion:The malignant risk nomogram prediction model constructed based on the independent risk factors mentioned above has a reliable clinical reference value, and it can assist sonographers in improving their diagnostic ability for NMLs.
基金项目
国家自然科学基金面上项目(82071934)
陕西省科技计划项目国际科技合作重点项目(2020KWZ-022)
陕西省高等教育教学改革研究重点项目(21JZ009)
空军军医大学临床研究项目(2021LC2210)
万方数据