燕山大学学报2024，Vol.48Issue(1)：86-94,9.DOI:10.3969/j.issn.1007-791X.2024.01.010
基于超声图像评估甲状腺和乳腺病变的通用计算方法
A general computational method of assessing thyroid and breast lesions based on ultrasound images
摘要
Abstract
Thyroid and breast nodules are common diseases that affect women's physical and mental health.In order to explore their individual differences and common features on ultrasound images,a general computational method is proposed to identify thyroid and breast nodules.The ultrasonic images are first decomposed into four sub-images by wavelet packet transform,and the gray level co-occurrence matrix is used to extract texture features of three sub-images without high-frequency noise.Effective features screened by the max-relevance and min-redundancy algorithm were respectively input four types of machine learning models to classify thyroid,breast benign and malignant nodules.The proposed method is applied to thyroid and breast ultrasound images from different platforms.The randomized cross-validation results show that the binary classification achieves the AUC of 0.88～0.99 and the accuracy(ACC)of 0.84～0.98,superior to the existed results,and the four classification achieves the AUC of 0.95～0.97 and the ACC of 0.88～0.92,which are better than that of ResNet50.Moreover,the four models trained on thyroid(breast)images are able to cross-identify benign and malignant breast(thyroid)nodules effectively,further validating the stability and generality of the proposed method.Meanwhile,the T-test results indicate that there are significant texture differences between approximate sub-images of the ultrasound images of malignant thyroid and breast nodules,and their vertical detail sub-images exhibit 6 potential common features.关键词
甲状腺/乳腺/良性结节/恶性结节/机器学习/超声图像Key words
thyroid/breast/benign nodule/malignant nodule/machine learning/ultrasound image分类
信息技术与安全科学引用本文复制引用
安俊达,李玉双..基于超声图像评估甲状腺和乳腺病变的通用计算方法[J].燕山大学学报,2024,48(1):86-94,9.基金项目
河北省自然科学基金资助项目(A2020203021) （A2020203021）
河北省引进留学人员资助项目(C20200365) （C20200365）
