Research On Detection Method Of CT Images Lesion Of Hepatic Cystic Echinococcosis

Objective:Apply the object detection method based on deep convolutional neural network to CT image processing of cystic liver echinococcosis,research the effect of activation function,regularization and Batch_Size on object detection performance,realize automatic detection of liver echinococcosis lesions,and analyze separately the features of the lesion area,thereby improving the automatic acquisition of effective information of image images,improving the diagnostic efficiency of imaging equipment,and at the same time,improving the development of a computer-aided diagnosis system for liver hydatid disease.Methods: In order to detect the exact location of the cystic hepatic hydatid disease lesion area in the two-dimensional CT image faster and more effectively,this study builds a deep learning single-stage object detection SSD model.Firstly,CT images of single-cystic and multi-cystic hepatic hydatid were collected,the data set was amplified by data enhancement technology to obtain 3000 images,and the amplified images were normalized by gray size,median filter to remove dryness and image enhancement Wait for pretreatment.Use the LabelImg image annotation tool to annotate the data to make a liver hydatidosis data set.In view of the problems existing in the test results of the SSD original model after initial training,this study improves the model’s activation function and regularization method to further optimize the model and improve model detection performance.First,the improved method compares the three activation functions of Tanh,ReLU and ELU,and studies the effect of different activation functions on the model performance.Then use the L1 and L2 methods to train the model and detect its performance,and study the impact of regularization methods on the model performance.Finally,under the condition that other variables are fixed,set the Batch_Size values to 8,16,and 32 to train the model,and explore the impact of these three Batch_Size values on the model detection performance.Finally,the model built by all theimproved methods is used for training,the lesion detection ability of the initial model and the improved model is compared,and the mean average precision mAP(AP_VOC07)evaluation index is used to evaluate the model lesion detection performance.Results:(1)After initial training using the original SSD model,the test was performed on the VOC test data set of liver echinococcosis,and the model test mAP value was 62.0 %.(2)The mAP values tested using Tanh,ReLU and ELU activation functions were 53.8 %,77.9 % and 83.2 %,respectively.(3)Using two regularization methods,L2 and L1,the model mAP values were 77.9 % and 81.2 %,respectively.(4)When Batch_Size is set to 32,16,and 8,the model test mAP values are 62.0 %,75.2 %,and 77.9 %,respectively.(5)The final training set Batch_Size to 8,combined with the improved model of ELU activation function and L1 regularization method,the test mAP value is 84.4 %.Conclusion:This study has been verified through a large number of experiments.In the detection of cystic liver hydatidosis CT image lesions,the ELU activation function,L1 regularization method and Batch_Size set to 8 are used to optimize the detection performance of the model.The improved model has improved detection accuracy,and it has better lesion detection ability.At the same time,this research realized the design and development of the interface of CT image lesion detection system for Hepatic Cystic Echinococcosis,which laid the foundation for the development of computer-aided diagnosis system of hepatic hydatid disease.