Study On Segmentation And Detection Method Of Tumor Medical Images Based On CNN

Cancer is a major public health problem in the world,and it tends to become the leading cause of human death.As a commonly used method of treatment and detection of treatment response in the medical field,medical images have become the main technical means for clinical diagnosis and treatment of cancer.With the advancement of medical technology,the efficiency of medical image generation has increased,but the limitations of traditional image processing methods limit the efficiency and accuracy of automated medical image processing.The advantages of convolutional neural networks for natural image processing make their applications in the medical field attract much attention.Using convolutional neural networks to process medical images can improve processing efficiency and processing accuracy.Based on convolutional neural networks,a liver medical image segmentation scheme based on joint loss function and a circulating tumor cell detection scheme fused with large-scale feature detection are proposed.The specific content is as follows:1)The current situation of medical image processing based on convolutional neural network is reviewed.First,this thesis analyzes the characteristics of medical image and its processing difficulties,and then the traditional medical image processing and medical image processing based on deep learning are reviewed.Finally,the current research status of medical image segmentation and detection based on convolutional neural network is explained separately.This thesis points out shortcomings in the existing medical image segmentation and detection research,and the research motivation of this thesis is clarified.2)Aiming at the problem of inaccurate segmentation of liver contour and small target regions in existing liver medical image segmentation task based on convolutional neural network,a liver medical image segmentation method based on the joint loss function is proposed.First,in order to improve the network segmentation performance for object edges and small regions,distance-based loss functions and region-based loss functions are combined to propose a liver medical image segmentation network based on joint loss function.Then,in order to make the joint loss function perform best effect,the weight in joint loss function is discussed,and a two-stage training strategy is proposed according to the characteristics of two loss function.Finally,the method proposed in this thesis is verified through experiments.The proposed joint loss function has increased the accuracy of liver segmentation by 1.2%and 1.9%on the two public data sets,and the accuracy of tumor segmentation has increased by 6.5%and 5.9%.The proposed training strategy improves liver segmentation accuracy by 1.4%and 1.1%,and tumor segmentation accuracy by 2.2%and 4.9%,respectively.The results of this article are better than other related studies.3)In response to the requirement of rapid and accurate detection of circulating tumor cells,this thesis proposes a circulating tumor cell detection scheme that extends feature scales.First,in order to enrich cell nucleus image features,a cell nucleus segmentation method with signal distribution feature is proposed.Then,in order to improve the speed and accuracy of signal point detection,a YOLO-V3-MobileNetL signal detection method was proposed based on the idea of multi-scale feature detection.Finally,based on the proposed cell nucleus segmentation and signal point detection methods,an automated circulating tumor cell detection process is designed.This thesis verifies through experiment that the method proposed is better than original method.Overall detection accuracy is increased by 3.15%,and detection speed is increased by 20%.