Research And Application Of Liver Cancer Prediction Algorithm Based On Contrast-enhanced MRI

Hepatocellular carcinoma(HCC)is the most common primary liver cancer and is the fourth leading cause of cancer death worldwide.Hepatectomy is one of the most effective treatments for hepatocellular carcinoma However,patients with hepatocellular carcinoma who undergo hepatectomy still have a recurrence rate of up to 70%.One important reason is that microvascular invasion may develop in patients with hepatocellular carcinoma.Microvascular invasion has very important guiding significance for patients with hepatocellular carcinoma undergoing surgery.In traditional medicine,the gold standard for diagnosing microvascular invasion phenomenon is to examine pathological images.Pathological images can only be obtained by postoperative tumor sampling and sectioning,and the results of whether microvascular invasion occurs after surgery have lost the timeliness of guiding surgery.To solve this problem,current studies have largely explored the use of contrast-enhanced MRI(magnetic resonance imaging)to predict microvascular invasion in patients with hepatocellular carcinoma before surgery.Enhanced MRI are also referred to as multi-sequence images in this thesis.At present,the study of microvascular invasion of hepatocellular carcinoma mostly uses multi-sequence images,and this kind of study generally uses multi-sequence image feature fusion model,which has a certain improvement in the prediction effect compared with the single sequence prediction model.However,these multi-sequence image feature fusion models often simply concatenation the image features extracted from different sequences,and do not fully utilize the features of multiple sequences as prior knowledge and combine them into the model,which limits the improvement of the prediction effect of the model.According to this research status,this thesis makes the following research and application of the algorithm for liver cancer prediction based on contrast-enhanced MRI.(1)In this thesis,a convolutional neural network model based on self-attention mechanism and image pyramid method is proposed.This model can make full use of the difference and correlation information between multi-sequence MRI to improve the prediction effect of the model.The model includes the multi-sequence image difference extraction pyramid module and the multi-sequence image correlation attention module proposed in this thesis.Among them,the multi-sequence image difference extraction pyramid module can extract the difference information between different sequences from multi-scale,while the multi-sequence image correlation attention module extracts the correlation information between multi-sequence images using the similarity calculation in the self-attention mechanism.The comparative experiment and ablation experiment in this thesis also showed that the prediction effect of the model was further improved on the basis of the multi-sequence fusion model.(2)In this thesis,a microvascular preoperative prediction assistant system for hepatocellular carcinoma was designed and implemented.The system is built with the convolutional neural network model based on self-attention mechanism and image pyramid method proposed in this thesis,which can realize one-click prediction of microvascular invasion after importing multi-sequence image data.In addition,the system also has the function of viewing multi-sequence images and their corresponding tumor segmentation markers,as well as the database function including image data,basic patient information,clinical information and other preoperative prediction information.These functions can help users to better collect and organize the information of patients with hepatocellular carcinoma,and assist users in preoperative prediction of microvascular invasion.