异构多分支网络超声图像自动诊断方法

Automatic Diagnosis Method of Ultrasound Image Based onHeterogeneous Multi-Branch Network

超声(US)是乳腺结节的主要影像学检查和术前评估手段之一,然而由于其在良性和恶性乳腺结节图像上的相似表现形式,使得诊断结果在很大程度上依赖于医生的经验判断,以计算机辅助手段帮助医生提高诊断准确率逐渐成为当前的热点。该文提出了一种用于乳腺超声图像的良性和恶性分类的异构多分支网络(HMBN),该网络同时使用了图像信息(超声图像及造影图像)和非图像信息(包括患者年龄和其他6个病理特征)。还提出了一种适合该异构多分支网络的混合型损失函数,在附加角边距损失的基础上应用了最小超球面能量来提升分类精度。实验结果表明,在本文收集的包含1303个病例的乳腺超声数据集上,该文提出的异构多分支网络的分类精度为92.41%,相较于具有5年资历的医师的平均诊断准确性提升了7.11%,相较于其他最新研究成果诊断准确性名列前茅。说明通过将医学知识纳入优化过程,以及将超声造影和非图像信息添加到网络中,可以在很大程度上提高乳腺超声诊断的准确性和鲁棒性。

Objective Ultrasound(US)is one of a primary imageological examination and preoperative assessment for breast nodules.However,in the field of ultrasound diagnosis,it relies heavily on the experience of physicians due to the overlapping image expression of benign and malignant breast nodules.Computer-aided medical diagnosis has gradually become a hot spot of current research.In this paper,a heterogeneous multi-branch network(HMBN)is presented for benign and malignant classification of the breast ultrasound images.In HMBN,the image information includes ultrasound images and contrast-enhanced ultrasound(CEUS)images while nonimage information includes patients’age and other six pathological features.On the other hand,a fusion loss function suitable for this heterogeneous multi-branch network is also proposed.This loss function uses the minimum hyperspherical energy(MHE)based on additive angular margin loss to improve the classification accuracy.Experimental results show that on the breast ultrasound data set of 1303 cases collected,the classification accuracy of the proposed heterogeneous multi-branch network is 92.41%,which is 7.11%higher than the average diagnostic accuracy of doctors with five years of experience,and ranks among the best in diagnostic accuracy in comparison with other latest research results.It is proved that the accuracy and robustness of breast diagnosis are greatly improved by incorporating medical knowledge into the optimization process and adding contrast-enhanced ultrasound images and non-image information to the network.