A Kind Of Research On Lesions Detection In Liver CT Images Based On Convolutional Neural Network

Deep learning is a research hotspot in the field of artificial intelligence.Target detection algorithm based on deep convolutional neural network is an important branch of deep learning.CT image lesion detection is an important way of computer aided medical care.Liver focal lesions pose great challenges for the detection of liver lesions due to the complexity of the background liver tissue and the diversity of lesions.Therefore,how to improve the detection rate of lesions is one of the researches on the detection of lesions in liver CT images.Focus and difficulty.This paper firstly generates a training and test data set based on the pictures of liver lesions marked by professional doctors in the imaging department,and then analyzes and builds the generalized Faster R-CNN.Then use the characteristics of CT image data of liver lesions,combined with lesion annotation information of different geometric locations,reasonable data interface format and training and test dataset division scheme,based on Faster R-CNN to establish corresponding liver lesion detection single classification and multi-classification Initial model.Because the detection effect of the initial model lesion is not ideal,the image preprocessing method,the feature extraction network,the value of the balance parameter in the loss function,and the ratio of the positive and negative samples of the target candidate frame are adjusted.In particular,the K-means is also used in this paper.The clustering algorithm generates and adjusts the number and size of different standard frames,and obtains the corresponding single-class and multi-class lesion detection models after optimization.Finally,numerical experiments were carried out to verify that the designed liver lesion detection model has higher average precision and applicability of multiple diseases.