Research On Detection And Classification Of Liver Lesions Based On Multi-phase CT Images

Worldwide,cancer has become the leading cause of death,with liver cancer ranking fourth.Early diagnosis and treatment of cancer are the main measures to reduce cancer mortality.One of the goals of computer-aided diagnosis(CAD)system is to realize the early diagnosis of cancer based on medical images,and it has always attracted the attention of academia and industry.The CAD system contains two important steps: lesion detection and lesion classification.Therefore,it is of great clinical application value to study accurate liver lesion detection and classification methods based on multiphase CT images.With the rapid development of artificial intelligence technology,deep learning models based on convolutional neural networks(CNN)have made breakthrough progress in many computer vision tasks.In recent years,many improved models based on CNN have also been widely used in CAD systems.However,the shape and size of the lesions in the multi-phase liver CT images vary greatly,various morphological changes of focal liver lesions,different types may have same features and the enhancement mode between the multi-phase images is difficult to describe.The above difficulties make the accurate detection and classification of liver lesions with great challenges.Therefore,this paper focuses on accurate liver lesion detection and classification algorithms.The main work is as follows:(1)Proposed the multi-stream scale-insensitive liver lesion detection algorithm based on multi-phase CT images.This paper proposes the multi-stream scale-insensitive liver lesion detection algorithm,which aim to tackle the problem of large changes in liver lesion size in multiphase CT images.The algorithm uses a U-shaped network structure,fuses visual information of different resolutions,and combines the Link mechanism to deal with the overlap between different lesion instances,and uses a convolutional recurrent neural network to extract inter-phase enhancement modes.Finally,the experimental results show that the detection algorithm proposed in this paper can significantly improve the detection effect of multi-phase liver lesions.When the IoU threshold is set to 0.5,the average detection accuracy(AP)reaches 77%,which is better than other existing methods.At the same time,our method of extracting inter-phase enhancement mode through recurrent neural network can be generalized to other application fields,which has certain generality and universality.(2)Proposed the combining global and local pathways liver lesion classification algorithm based on multi-phase CT images.When diagnosing liver lesions,doctors often shrink the liver lesion image to obtain local and global information at the same time,and the doctor will switch among phases to obtain the difference in the performance of the lesion among phases to avoid heterogeneous diseases.Inspired by the phenomenon,this paper first proposes a liver lesion classification algorithm combining local and global information.The algorithm simultaneously inputs a patch representing local information and a region of interest(RoI)representing global information into the network at the same time,and uses different stream to process local information and global information,respectively.Secondly,this paper uses a fully-connected long-short-term memory network to extract inter-phase enhancement patterns.Finally,the experimental results show that the multiphase liver lesion classification algorithm proposed in this paper can fuse local,global,enhanced mode and other relevant diagnostic information,and it is effective.The classification accuracy rate is about 88%,which is superior to other existing methods.