Research And Application Of Liver Lesion Detection Method Based On Faster R-CNN

In recent years,the amount of liver cancer patients and mortality of liver cancer has increased gradually,which has been rowed in the third place among cancer deaths.Accurate detection of liver tumors on abdominal CT can help doctors evaluate and diagnose better.With the continuous advancement of artificial intelligence algorithms,deep learning algorithms are widely utilized in medical image detection tasks.Through the study of detection algorithms based on deep learning,the automatic labeling of liver tumors in abdominal CT can be achieved to assist doctors in diagnosis.However,the problem of small resolution and variable scales is still hard to solve which is impreceptible.In cope with the above problems,this article has carried out a research on liver tumor detection methods based on Faster R-CNN.The details are as follows:1.The proposed Con A-FPN-based algorithm for liver tumor detection is as follows:first,the feature extraction network in Faster R-CNN is replaced using a feature pyramid fused with Res Net and attention mechanism;then,the fused features are used to solve the problem of information loss of higher-level module channels in the feature pyramid,and the feature fusion brought about by feature fusion is solved by adding the CAG attention mechanism to enhance the model’s detection capability for small tumors.The problem of blending is solved by adding the CAG attention mechanism to enhance the detection ability of the model for small tumors;finally,transfer learning and data augmentation are used to enhance the detection ability and generalization ability of the model on small data sets.The experimental results show that the m AP of the network can reach 0.874,which effectively improves the detection accuracy.2.A liver tumor detection method based on U-shaped auxiliary network is proposed.This method applies a U-net network-based structure instead of the DB dual backbone structure in CBNet,which marked reduce computation and parameter,meanwhile ensuring improved performance via fusing differnet-level feature.Secondly,Io U loss is utilized to replace Faster R-CNN’s original smooth L1 loss to optimize the bounding box regression process.Experimental results and visualization results demonstrate that this method can not only improves the accuracy,but also makes the position of the bounding box more accurate.3.For the two methods mentioned above,the liver tumor detection system is proposed to detect liver tumors using a deep learning-based approach,which can reduce the rate of false detections and missed detections caused by subjective evaluations and overwork of physicians while reducing the burden on physicians.