摘要
针对医学图像分辨率小、边缘模糊、感兴趣区域(ROI)边界不明显造成的分割不准确性问题,提出一种新型MobileUnet网络的肺结节图像分割方法。该方法首先使用MobileNet中bneck模块替换Unet网络的下采样部分,并对输入图像进行特征提取;然后,将下采样提取的特征按照Unet网络连接方式融合到上采样部分;最后,利用训练好的网络得到分割结果。实验利用采集的肺结节数据集对MobileUnet进行了训练和验证。结果表明:新型MobileUnet网络能够更加准确地分割出肺结节位置,分割准确率从原有Unet网络的85.00%提高至90.00%,同时相似系数(F_(1))仍然稳定在89.98%,证明该方法的有效性。
To improve the segmentation inaccuracy caused by low resolution of medical images,blurred edges,and insignificant region of interest(ROI)regions,a new method for the segmentation of lung nodules based on Mobile-Unet network was proposed.Firstly,the bneck module in MobileNet was used to replace the downsampling part of the Unet network,and the features of the input image was extracted.Then,the features extracted by the downsampling were fused to the upsampling part according to the Unet network connection.Finally,the trained network was applied to the segmentation task.Results showed that the proposed Mobile-Unet network could segment the lung nodules more accurately as it increased the accuracy from 85.00%for the original Unet network to an improved level about 90.00%,while the similarity coefficient(F_(1))could still kept at 89.98%,which proved the effectiveness of the method.
作者
陈铭
梅雪
朱文俊
周颖
张梦怡
冯李航
CHEN Ming;MEI Xue;ZHU Wenjun;ZHOU Ying;ZHANG Mengyi;FENG Lihang(College of Electrical Engineering and Control Science, Nanjing Tech University, Nanjing 211800, China)
出处
《南京工业大学学报(自然科学版)》
CAS
北大核心
2022年第1期76-81,91,共7页
Journal of Nanjing Tech University(Natural Science Edition)
基金
国家自然科学基金(61803198)
江苏省自然科学基金(BK20180701)。