北京生物医学工程

超声图像融合专家知识的ＥｆｆｉｃｉｅｎｔＮｅｔ卵巢包块良恶性诊断方法

EfficientNet diagnosis method of benign and malignant ovarian masses based on ultrasound images incorporating expert knowledge

作者：赵君逸叶萍石思远杨洋徐明杰常兆华

单位：上海理工大学健康科学与工程学院（上海 200093）<br />上海微创医疗器械（集团）有限公司（上海 201203）<br />通信作者：常兆华，教授。E-mail：m8090313@126.com

关键词： EfficientNet；卵巢包块；特征融合；计算机辅助诊断；良恶性分类

分类号：R318.04

出版年·卷·期（页码）：2023·42·1（27-32）

摘要：

目的提出一种基于深度网络特征融合的分类方法，以提高良恶性分类的准确率，达到辅助医生提高术前诊断卵巢包块良恶性准确率的目的。方法纳入深圳市人民医院以943幅经活检、手术病理等证实的患者术前卵巢超声图像，按照6:2:2的比例随机设置训练集、验证集和测试集。首先，提取医生勾画的感兴趣区域（region of interest，ROI）即包块图，用微调后的EfficientNet网络提取其深度特征；然后用基于Chan-Vese模型的水平集方法得到包块边缘轮廓图，再用微调后的EfficientNet网络提取其深度特征；接下来将包块图的深度特征和边缘轮廓图的深度特征分别归一化后并拼接为融合特征；最后，将融合特征输入到全连接层分类器中，将超声图像分为良恶性。结果本文提出的超声图融合专家知识的EfficientNet卵巢包块良恶性诊断方法在测试集上的准确度、特异度、敏感度和曲线下面积分别为 0. 81,0. 78,0. 88,0. 91,全部优于当前主流的深度学习方法。结论该特征融合网络能够取得较好的分类效果,一定程度上能够为临床诊断卵巢包块的良恶性提供参考。

Objective To propose a classification method based on deep network feature fusion for improving the accuracy of benign and malignant classification, and to assist doctors improve the accuracy of preoperative diagnosis of benign and malignant ovarian masses. Methods The preoperative ovarian ultrasound images of 943 patients confirmed by biopsy and surgical pathology in Shenzhen People's Hospital are included, and the training set, validation set and test set are randomly set according to the ratio of 6:2:2. First, extract the region of interest (ROI) delineated by the doctor, that is, the block map, and use the fine-tuned EfficientNet network to extract the depth features of the block map; then use the level set segmentation method based on the Chan-Vese model to obtain the block. Edge contour map, and then use the fine-tuned EfficientNet network to extract its depth features; next, the depth features of the block map and the edge contour map are respectively normalized and spliced into fusion features; finally, the fusion features are input into the fully connected layer classifier, ultrasound images are classified into benign and malignant. Results The accuracy, specificity, sensitivity and area under curve of the EfficientNet method for the diagnosis of benign and malignant ovarian masses based on the fusion of expert knowledge of ultrasound images proposed in this paper were 0. 81,0 on the test set, respectively 0. 78, 0. 88, 0. 91, all of which are better than the current mainstream deep learning methods. Conclusions The network can achieve good classification effect, and to a certain extent can provide a reference for clinical diagnosis of benign or malignant ovarian masses.

参考文献：

[1] Stewart C, Ralyea C, Lockwood S. Ovarian cancer: an integrated review[J]. Seminars in Oncology Nursing, 2019, 35(2): 151-156.
[2] Sadowski E A, Thomassin-Naggara I, Rockall A, et al. O-RADS MRI risk stratification system: guide for assessing adnexal lesions from the ACR O-RADS committee[J]. Radiology, 2022: 303(1):35-47.
[3] Acharya UR, Saba L, Molinari F, et al. Ovarian tumor characterization and classification: A class of GyneScan? systems[C]//2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society.San Diego: IEEE, 2012: 4446-4449.
[4] Acharya UR, Sree S V, Kulshreshtha S, et al. GyneScan: an improved online paradigm for screening of ovarian cancer via tissue characterization[J]. Technology in Cancer Research & Treatment, 2014, 13(6): 529-539.
[5] Khazendar S, Sayasneh A, Al-Assam H, et al. Automated characterisation of ultrasound images of ovarian tumours: the diagnostic accuracy of a support vector machine and image processing with a local binary pattern operator[J]. Facts, Views & Vision in ObGyn, 2015, 7(1): 7.
[6] Kiruthika V, Sathiya S, Ramya MM. Machine learning based ovarian detection in ultrasound images[J]. International Journal of Advanced Mechatronic Systems, 2020, 8(2-3): 75-85.
[7] Zebari R, Abdulazeez A, Zeebaree D, et al. A comprehensive review of dimensionality reduction techniques for feature selection and feature extraction[J]. Journal of Applied Science and Technology Trends, 2020, 1(2): 56-70.
[8] Nixon M, Aguado A. Feature extraction and image processing for computer vision[M]. New York:Academic Press, 2019.
[9] Tan D, Chen L, Jiang C, et al. A circular target feature detection framework based on DCNN for industrial applications[J]. IEEE Transactions on Industrial Informatics, 2020, 17(5): 3303-3313.
[10] Souli S, Amami R, Yahia S B. A roBust pathological voices recognition system based on DCNN and scattering transform[J]. Applied Acoustics, 2021, 177: 107854.
[11] Jeyaraj PR, Nadar ERS. Computer-assisted medical image classification for early diagnosis of oral cancer employing deep learning algorithm[J]. Journal of Cancer Research and Clinical Oncology, 2019, 145(4): 829-837.
[12] Saleh A Y, Chin C K, Penshie V, et al. Lung cancer medical images classification using hybrid CNN-SVM[J]. International Journal of Advances in Intelligent Informatics, 2021, 7(2): 151-162.
[13] 邹奕轩, 周蕾蕾, 赵紫婷, 等. 基于卷积神经网络的甲状腺结节超声图像良恶性分类研究[J]. 中国医学装备, 2020, 17(3): 9-13.
Zou YX，Zhou LL，Zhao ZT，et al.Study on the classification of benign and malignant thyroid nodule in ultrasound image on the basis of CNNs[J]. China Medical Equipment, 2020, 17(3): 9-13.
[14] Zhang Y Z, Wang Z H, Zhang J, et al. Deep learning model for classifying endometrial lesions[J]. Journal of Translational Medicine, 2021, 19(1): 1-13.
[15] Zhang L, Huang J, Liu L. Improved deep learning network based in combination with cost-sensitive learning for early detection of ovarian cancer in color ultrasound detecting system[J]. Journal of Medical Systems, 2019, 43(8): 1-9.
[16] 程广文，漆玖玲，丁红. 超声医学在卵巢肿瘤诊断和化疗效果评估中应用的研究进展[J].复旦学报(医学版),2021,48(4):545-550.
Cheng GW，Qi JL，Ding H. Research progress on ultrasound medicine in diagnosis and evaluation of chemotherapy effect of ? ? ovarian tumor[J]. Fudan University Journal of Medical Sciences, 2021, 48(4): 545-550.
[17] 杨曜彰, 余金穗，陈智毅. 超声影像学在附件包块诊断中的应用[J]. 中国实用妇科与产科杂志, 2020, 36(12): 1153-1157.
Yang YZ，Yu JS，Chen ZY. Application of ultrasound imaging in the diagnosis of adnexal masses[J]. Chinese Journal of ? ? Practical Gynecology and Obstetrics, 2020, 36(12): 1153-1157.
[18] Chan TF, Vese LA. Active contours without edges[J]. IEEE Transactions on Image Processing, 2001, 10(2): 266-277.
[19] Tan M, Le Q. Efficientnet: Rethinking model scaling for convolutional neural networks[C]//International Conference on Machine Learning. Long Beach:PMLR, 2019: 6105-6114.
[20] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[C]//International Conference on Learning Representations.San Diego:OpenReviw,2015: 1-14.
[21] Huang G, Liu Z, Van Der Maaten L, et al. Densely connected convolutional networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. Hawaii:IEEE, 2017: 4700-4708.
[22] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. Las Vegas:IEEE,2016: 770-778.
[23] 焦蕊丽,郑华,徐佳圆, 等.三维超声联合HE4和CA125水平检测对卵巢良恶性肿瘤鉴别诊断价值[J].中国医学装备,2021,18(7):88-93.
Jiao RL，Zheng H，Xu JY，et al.The diagnostic value of 3D ultrasonography combined with HE4 and CA125 in identifying ? ? benign and malignant tumors of ovary[J]. China Medical Equipment, 2021,18(7):88-93.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录