计算机辅助多模态融合超声诊断乳腺良恶性肿瘤

目的 观察利用深度学习（DL）融合常规超声和超声弹性成像诊断乳腺良、恶性肿瘤的效能。方法 利用DL卷积神经网络（CNN）提取乳腺肿瘤超声灰阶与超声弹性特征，并进行多模态融合，评价融合弹性图像或弹性比值等不同信息方式对乳腺良、恶性肿瘤的诊断效能；绘制受试者工作特征（ROC）曲线，计算曲线下面积（AUC），评估多模态融合模型的诊断效能。结果 多模态融合模型鉴别乳腺良、恶性肿物的效能优于单模态常规超声或弹性模型，其中融合灰阶与弹性图像模型鉴别诊断效能优于融合灰阶与弹性比值模型，分类准确率达93.51%，敏感度为94.88%，特异度为92.25%，AUC达0.975。结论 计算机辅助多模态融合有助于提高超声对乳腺良、恶性肿瘤的诊断效能。

Computer-aided multimodal fusion ultrasonic diagnosis of benign and malignant breast tumors

Objective To observe the efficacy of fusion of conventional ultrasound and ultrasonic elastography using deep learning (DL) for diagnosis of benign and malignant breast tumors. Methods Convolutional neural network (CNN) was used to extract the features of ultrasonic grayscale images and ultrasonic elastic images, and multimodal fusion was performed. The performances of fusion of elastic images or elastic ratio data and other different information methods in diagnosis of benign and malignant breast tumors were evaluated, respectively. Receiver operating characteristic (ROC) curves were drawn, and the areas under the curves (AUC) were calculated to further assess the efficacy of the models.Results The multimodal fusion models were superior to single modal of conventional ultrasound or elastic model for differentiating benign and malignant breast tumors. The fusion model of gray-scale image and elastic image was superior to that of gray-scale image and elastic ratio, with classification accuracy of 93.51%, sensitivity of 94.88% and specificity of 92.25%, and the AUC was 0.975. Conclusion Computer-aided multimodal fusion could help to increase the efficacy of ultrasonic diagnosis of benign and malignant breast tumors.