基于DYOLO神经网络的超声图像肾脏检测

为便于慢性肾脏疾病的计算机辅助诊断,提出一种基于DYOLO神经网络学习模型的自动超声图像肾脏检测方法.将YOLOv3和可变形卷积网络集成在一个端到端学习框架中,使得DYOLO可根据肾脏的大小和形状自适应调节接收域,以适应肾脏的各种纹理特征形变,实现临床超声图像中肾脏的自动检测.在自制KidneyDetec超声图像肾脏检...

Kidney Detection Using Ultrasound Image Based on DYOLO Neural Network

In order to facilitate the Computer Aided Diagnosis(CAD) of Chronic Kidney Disease(CKD),a kidney detection using ultrasound image based on DYOLO neural network model is proposed.The method integrates YOLOv3 and Deformable Convolutional Network(DCN) into an end-to-end learning framework,making DYOLO adaptively adjust the receiving area according to the size and shape of the kidney to adapt to various texture feature deformations,and realizes automatic kidney detection in clinical ultrasound images.Experimental results on a self-made ultrasound image-based kidney detection dataset,KidneyDetec,show that the mean Average Precision(mAP) of the proposed method reaches 89.6% when the image size of the DYOLO input is 416×416 pixels,and 90.5% when the image size of the DYOLO input is 608×608 pixels.Compared with deep learning-based target detection methods,the proposed method has higher detection speed and accuracy,displaying excellent applicability to the early diagnosis of chronic kidney diseases.