Research On Detection And Segmentation Of Specific Organs In Medical CT Images Based On Deep Learning

In recent years,with the continuous increase of various types of data sets and the acceleration of computer hardware,the field of artificial intelligence has achieved unprecedented development,and deep learning as a core driving technology of artificial intelligence has a significant impact on human life.Medical image segmentation has important applications for intelligent medical treatment,and has gradually become a research hotspot in recent years.However,due to the particularity and limitations of medical images,medical image segmentation is more challenging than other computer vision tasks.Therefore,research on medical image segmentation is urgently needed.This paper takes convolutional neural networks as the research direction,and rationally designs and improves models to improve the detection and segmentation performance of medical images.The main research contents are as follows:1.Proposed a network model of bladder tumor detection based on Mask R-CNNAiming at the problems of cumbersome steps,low accuracy,and poor detection effect and efficiency in existing deep learning methods in traditional image detection methods,a detection network model based on Mask-RCNN is proposed.The detection network includes a proposed random weighting algorithm,which randomly assigns lightweight weights to ResNet to reduce the computational complexity,remove excessive redundant information from the network,and improve the computational efficiency.In addition,according to the characteristics of the bladder medical data set,reduce the number of network layers and reduce the risk of network overfitting.The experimental results show that: first,the comparison of different ResNet layers proves that reducing the number of network layers can effectively improve the detection accuracy and training speed of bladder tumors;second,the random weighting algorithm for bladder tumor detection accuracy and Training speed has improved.2.Proposed a CT image segmentation model based on multi-scale dilatedConvolution Network(Md-Net).The accurate segmentation of CT images is of great significance for clinical diagnosis.Due to the sensitivity and specificity of CT images and artifacts in CT imaging,the semantic information of the target is lost,which affects the segmentation performance.Therefore,a convolution of holes of different sizes is used to form a multi-scale feature pyramid to extract more semantic information and improve the segmentation performance of CT images.In addition,during the training process,a weighted Dice loss function is used to accelerate convergence.At the same time,bilinear interpolation and 1x1 convolution kernel are used to reduce the computational complexity.Experimental results show that the performance of Md-Net on the bladder dataset and lung dataset is better than the advanced models such as Unet,Unet ++,Mask R-CNN and CE-Net.