Research On Lymph Node Ultrasound Image Segmentation Algorithm Based On Deep Learning

Ultrasonography is the most commonly used examination method for lymph node diseases.To diagnose lymph node diseases by observing ultrasound images,it is first necessary to segment lymph nodes from ultrasound images,which is time-consuming and labor-intensive,and requires high experience of doctors.Therefore,automatic segmentation technology of lymph node ultrasound image is an important auxiliary means to assist doctors in the diagnosis of lymph node related diseases.However,there are two main problems in the current lymph node ultrasound image segmentation technology.The first problem is that the traditional image segmentation technology cannot extract the depth features of ultrasound images,so the segmentation accuracy is not ideal.The second problem is that the current classical network model based on deep learning is not designed for lymph node ultrasound image segmentation,which will also make the segmentation accuracy insufficient.In the thesis,we studied the automatic segmentation of lymph node ultrasound images based on deep learning,proposed a data pre-processing method for multimodal image fusion,designed a scheme for automatic segmentation of lymph node ultrasound images,and realized the accurate segmentation of lymph node ultrasound images.The main work and results of the thesis are as follows:1.To meet the high precision requirements of lymph node ultrasound image segmentation,a convolutional neural network model DMA-UNet for lymph node ultrasound image segmentation was designed.The original single 3*3 convolution was replaced by residual convolutional block and multi-scale void convolutional block in the coding and decoding structure,and a mixed attention mechanism was adopted in the jump long connection path.Experiments show that the Dice coefficient and IOU index of the proposed model are improved by 16%and 28%compared with the UNet network model,respectively.2.A multi-modal image fusion data preprocessing method is proposed to preprocess lymph node ultrasound images,which solves the problem of accurate segmentation of complex lymph node images.Based on DMA-UNet network model,the segmentation effect of multi-modal image fusion method is verified by experiments.The results show that the segmentation results of the proposed method improve the Dice coefficient and IOU index by nearly 3%and nearly 1%,respectively.3.An automatic segmentation system of lymph node ultrasound images with GUI interface is designed and implemented,which realizes the fast and accurate segmentation of lymph nodes,eliminates the tedious computer processing process,and improves the work efficiency of doctors.