Application And Research Of Placenta Image Segmentation Technology Based On Deep Learning

Medical images can help doctors quickly identify,diagnose and treat diseases in the human body,but relying solely on human labor to identify specific areas of the human body can consume a lot of doctors' time and energy.With the rise of computer vision in recent years,research on algorithms for recognition,detection,and segmentation of specific regions in medical images has made great progress.The use of specific algorithms can help doctors to quickly make judgments on medical images.Take placenta images as an example,the use of image segmentation technology can quickly help doctors to segment the placenta and the complex background environment,thus further help doctors to better analyze the condition.However,the existing traditional and deep learning model algorithms have the disadvantages of low segmentation accuracy and high model complexity,which will lead to inaccurate segmentation of placenta images and serious consequences of doctors' diagnosis errors.For the above problems,this thesis proposes a Dense Connection Based On Dense U-Net network model and a PSA-UNet network model based on attention mechanism from different principles,and the experiment has achieved a good segmentation effect on the placental dataset.The main tasks of this article are as follows:(1)This thesis medical datasets are organized to construct a placenta image dataset suitable for the image segmentation field.At the same time,in order to verify the superiority of Deep Learning in the field of semantic segmentation,an SVM classifier is used to perform image segmentation on placental images,and experiments prove that traditional machine learning algorithms perform poorly in placental image segmentation.(2)This thesis proposes a network model for Sense U-Net.Firstly,the defects of the traditional UNet network are analyzed,the loss function is optimized,and the binary cross-entropy loss function and the Dice loss function are mixed.The dense connection module is used to optimize the traditional U-Net network,improve the feature multiplexing rate,reduce the parameters in the network,and the dense connection structure makes the gradient transmission more fluent,makes the network easier to train,reduces the disappearance of the gradient,and improves the segmentation accuracy of the network model for the image.At the same time,the model pruning technology is used to compress the model,which reduces the number of parameters in the model by70% while ensuring that the accuracy of the image segmentation task is not significantly reduced.In the placental image segmentation experiment,Dense U-Net performed better than the traditional U-Net on all evaluation indicators,with pixel accuracy of0.8734,Dice coefficient of 0.8910,and IOU of 0.8112.(3)This thesis proposes a PSA-UNet(Placental segmentation based on SE and AG-UNet)network model.Compared with the traditional U-Net network model,the model combines the SE-Res Net module and the AG module,which enables the model to extract more valuable feature information on the placental dataset and compress the model size in combination with the model pruning technology.The experimental results on the placental dataset show that compared with the compression segmentation model based on Sense-UNet,the pixel accuracy rate increased by 2% without significantly increasing the memory occupancy of the model,and all indicators were improved to a certain extent,with pixel accuracy reaching 0.8927,Dice coefficient reaching 0.8946,and IOU reaching 0.8231,which had better performance.