Multi-organ Fusion And LightGBM Based Radiomics Algorithm And Its Applications In Esophageal Varices

Esophageal varices(EV)is one of the most common complication of portal hypertension in cirrhosis patients.Radiomics has been progressing remarkably for quantifying the state of diseases.At present,the golden diagnostic standard of EV mainly relies on gastroscopy in the clinic.However,the gastroscopy is invasive which may carry lots of uncomfortable reactions to patients.Therefore,it is of great significance to apply non-invasive medical imaging technology to predict the risk level of esophagus varices and assist the diagnosis of liver cirrhosis,which can not only reduce the pain of the patients but also be safety and repeatability.However,there are few studies on EV severity prediction by applying radiomics and machine learning.Besides,most of the existing methods apply only single organ(e.g.,liver or spleen)for radiomics feature extraction.Therefore,we apply a variety of machine learning methods to EV subjects based on radiomics.Moreover,we further propose multi-organ fusion models to improve EV risk diagnosis performance.The main research works are as follow:1.We collect liver cirrhosis data from different hospitals.The approach involves 188 patients,including 151 cirrhotic patients(84 patients with severe EV and 67 patients with mild or no EV)registered in Qilu Hospital of Shandong University from January 2018 to August2020,and 37 cirrhotic patients(20 patients with severe EV and 17 patients with mild or no EV)retrospectively registered in Jinan Central Hospital from May 2020 to June 2020.2.We propose an EV risk prediction algorithm based on Light Gradient Boosting Machine(LightGBM).In this paper,the esophageal region is firstly applied to EV risk diagnosis and the radiomics features of liver,spleen and esophagus are extracted after manual segmentation.Meanwhile,LightGBM is applied to EV risk diagnosis to establish classification model for the first time.We study the predictive performance of different organs based on LightGBM for EV.Experimental results demonstrate that the feature of esophageal region makes a greater contribution on the diagnosis of EV compared with spleen and liver.In addition,LightGBM has the best predictive performance compared with traditional methods.3.We propose an EV risk prediction algorithm based on multi-organ fusion models.In order to further discriminate the high-risk EV patients,we propose multi-organ fusion models based on liver,spleen and esophagus.Firstly,we propose multi-organ concatenated features model,which are generated by concatenating the features of three organs.Secondly,we propose multi-organ features linear fusion model,the features of the three organs are fused by linear combination where the weights are estimated by the feature distribution.Finally,we propose multi-organ classifier ensemble strategy by combining the prediction probability of each organ for final classification.Experimental results demonstrate that the prediction performance of the multi-organ concatenated feature proposed in our paper is significantly better than liver,spleen and esophagus.More importantly,the proposed multi-organ linear fusion feature and classifier ensemble models greatly improve the performance of EV risk diagnosis.In addition,LightGBM has the best predictive performance compared with traditional methods.