结合X线和超声的神经网络在乳腺检查的应用(1)
摘 要：目的 探讨神经网络方法在鉴别乳腺X线和超声检查中良恶性病变的意义。方法 对75例乳腺疾病患者的X线钼靶及超声检查的数据用NN进行分析，随机选择30例样本作为训练样本，组成训练集，其余样本组成测试集。建立NN模型，分析神经网络模型的诊断结果。结果 75例患者中，手术与病理证实乳腺恶性病变44例，乳腺良性肿瘤或肿瘤样病变31例，钼靶X线诊断的特异度、敏感度及诊断正确率分别为90.32%、88.64%和89.33%，B超诊断的特异度、敏感度及诊断正确率分别为87.09%、86.36%和86.67%，X线钼靶和B超对比，差异无统计学意义(P>0.05)。而BP网络的特异度为95.45%，敏感度为95.65%，总正确率为95.56%，高于X线钼靶和B超，差异具有统计学意义(P<0.05)。结论 神经网络具有人脑的学习、不断进步的优点，又比人脑客观，结合X线钼靶和B超检查的神经网络在判断乳腺良恶性病变性质方面有一定的应用价值。
关键词：神经网络；X线钼靶摄影；超声检查；乳腺病变
, http://www.100md.com
中图分类号：R737.9 文献标识码：A DOI：10.3969/j.issn.1006-1959.2018.13.003
文章编号：1006-1959(2018)13-0009-04
Abstract：Objective To explore the significance of neural network in the differential diagnosis of benign and malignant lesions in X-ray mammography and ultrasonography.Methods The data of X-ray mammography and ultrasonography in 75 patients with breast disease were analyzed by NN.30 samples were randomly selected as training samples to form a training set，and the remaining samples were composed of test sets.Establish a NN model and analyze the diagnosis results of the neural network model.Results Among the 75 cases，44 cases of malignant breast lesions，31 cases of benign breast tumor or tumor like lesion were confirmed by operation and pathology.The specificity，sensitivity and diagnostic accuracy of molybdenum target X-ray diagnosis were 90.32%，88.64% and 89.33% respectively.The specificity， sensitivity and diagnostic accuracy of B ultrasonic diagnosis were 87.09%，86.36% and 86.67%， respectively，there was no significant difference between X-ray mammography and B-ultrasound(P>0.05).The specificity of BP network was 95.45%，the sensitivity was 95.65%，and the total correct rate was 95.56%，which was higher than that of X-ray mammography and B-mode ultrasound，the difference was statistically significant(P<0.05).Conclusion Neural network has the advantages of learning and continuous improvement of human brain.It is more objective than human brain.The neural network combined with X-ray mammography and B-ultrasound has certain application value in judging the nature of benign and malignant breast lesions.
, 百拇医药
Key words：Neural network；X-ray mammography；Ultrasonography；Breast lesions
乳腺癌(mammary cancer)的發病率在全球范围内呈上升趋势，位居女性恶性肿瘤第一位。近年来，女性乳腺癌的发病率和死亡率呈持续上升趋势，并且有年轻化趋势[1]。乳腺X线钼靶摄影是最基本的检查方法，特别是对于发现细小钙化最有优势[2，3]；而乳腺超声检查具有简便、无创、无辐射及可反复等特点，特别是对X线检查阴性的致密性乳腺患者的检查较具优势。但受到乳腺良恶性肿块影像特征影响，这两种影像检查方法单独检查有时定性非常困难，易出现误诊或者漏诊的情况，可能耽误患者治疗，尽管有作者联合超声和X线钼靶检查，能够有效提高检查水平[4，5]，但两种检查结论经常矛盾，存在分析困难的问题。为了突破依赖于主观判断的传统的乳腺病变影像诊断模式，我们尝试应用基于深度学习反向传输神经网络方法回顾性分析本院2010年3月～2016年6月进行X线钼靶和B超检查的乳腺病变，目的在于探讨神经网络方法在鉴别乳腺影像检查中良恶性病变的意义。, http://www.100md.com(张琦 杜丽娟 唐震)
关键词：神经网络；X线钼靶摄影；超声检查；乳腺病变
, http://www.100md.com
中图分类号：R737.9 文献标识码：A DOI：10.3969/j.issn.1006-1959.2018.13.003
文章编号：1006-1959(2018)13-0009-04
Abstract：Objective To explore the significance of neural network in the differential diagnosis of benign and malignant lesions in X-ray mammography and ultrasonography.Methods The data of X-ray mammography and ultrasonography in 75 patients with breast disease were analyzed by NN.30 samples were randomly selected as training samples to form a training set，and the remaining samples were composed of test sets.Establish a NN model and analyze the diagnosis results of the neural network model.Results Among the 75 cases，44 cases of malignant breast lesions，31 cases of benign breast tumor or tumor like lesion were confirmed by operation and pathology.The specificity，sensitivity and diagnostic accuracy of molybdenum target X-ray diagnosis were 90.32%，88.64% and 89.33% respectively.The specificity， sensitivity and diagnostic accuracy of B ultrasonic diagnosis were 87.09%，86.36% and 86.67%， respectively，there was no significant difference between X-ray mammography and B-ultrasound(P>0.05).The specificity of BP network was 95.45%，the sensitivity was 95.65%，and the total correct rate was 95.56%，which was higher than that of X-ray mammography and B-mode ultrasound，the difference was statistically significant(P<0.05).Conclusion Neural network has the advantages of learning and continuous improvement of human brain.It is more objective than human brain.The neural network combined with X-ray mammography and B-ultrasound has certain application value in judging the nature of benign and malignant breast lesions.
, 百拇医药
Key words：Neural network；X-ray mammography；Ultrasonography；Breast lesions
乳腺癌(mammary cancer)的發病率在全球范围内呈上升趋势，位居女性恶性肿瘤第一位。近年来，女性乳腺癌的发病率和死亡率呈持续上升趋势，并且有年轻化趋势[1]。乳腺X线钼靶摄影是最基本的检查方法，特别是对于发现细小钙化最有优势[2，3]；而乳腺超声检查具有简便、无创、无辐射及可反复等特点，特别是对X线检查阴性的致密性乳腺患者的检查较具优势。但受到乳腺良恶性肿块影像特征影响，这两种影像检查方法单独检查有时定性非常困难，易出现误诊或者漏诊的情况，可能耽误患者治疗，尽管有作者联合超声和X线钼靶检查，能够有效提高检查水平[4，5]，但两种检查结论经常矛盾，存在分析困难的问题。为了突破依赖于主观判断的传统的乳腺病变影像诊断模式，我们尝试应用基于深度学习反向传输神经网络方法回顾性分析本院2010年3月～2016年6月进行X线钼靶和B超检查的乳腺病变，目的在于探讨神经网络方法在鉴别乳腺影像检查中良恶性病变的意义。, http://www.100md.com(张琦 杜丽娟 唐震)