Study On Clinical Decision Support Of Liver Cancer Based On Imaging Data Mining

[Objective]The present study focuses on the pracitcal problems in the clinical decision support of liver cancer,we aimed to improve the ability of clinical decision support and the efficiency of patient management via machine learning.First,we aimed to investigate the performance of the B-mode ultrasound-based radiomics approach to differentiate primary liver cancer from metastatic liver cancer,and to promote the predictive accuary compared to traditional methods.Second,we aimed to investigate whether contrast enhanced CT-based radiomics signatures could be applied to preoperatively predict pathological grades of HCC and the accuary.Third,we aimed to develop a contrast enhanced CT-based radiomics model using a machine learning method and assess its efficacy of preoperative prediction for the early recurrence of HCC,and to enhance the predictive ability of risk stratification.[Methods]The present study comprehensively applied various methods such as machine learning,image processing,and data mining to construct radiomics models aiming to explore the underlying relationship between radiomics signatures and biological behavior of tumors.First,data of 114 consecutive histopathologically confirmed patients with liver cancer were retrospectively analyzed.All patients underwent liver ultrasonography within 1 week before hepatectomy or fine-needle biopsy.The liver lesions were manually segmented by two experts using ITK-SNAP software,and features were extracted by Pyradiomics.Then,the dimensions of radiomics features were reduced by Lasso method.Finally,five machine learning methods were employed to establish the model.Second,data collected from 297 consecutive subjects with HCC were allocated to training dataset and test dataset.Manual segmentation of lesion sites was performed with ITK-SNAP,the features were extracted by the Pyradiomics,and radiomics signatures were synthesized using RFE method.The prediction models for pathological grading of HCC were established by XGBoost.Third,a total of 297 patients confirmed with HCC were collected,all patients were followed up at least 1 year.Manual segmentation of lesion sites was performed with ITKSNAP,and the radiomics features were extracted by the pyradiomics platform.The prediction models for early recurrence of HCC were established by using deep learning.[Results]First,five machine learning classifiers were able to differentiate primary liver cancer from metastatic liver cancer.LR outperformed other classifiers,with the AUC 0.816 ± 0.088.Second,the radiomics signatures were found highly efficient for machine learning to differentiate high-grade HCC from low-grade HCC.The radiomics signatures were applied in association with clinical factors to train a machine learning model,the performance of the model remarkably increasedwith AUC of 0.8014.Third,the radiomics signatures were found highly efficient for deep learning to predict the early recurrence of HCC.When arterial phase and venous phase images were applied in association with clinical factors to train a deep learning model,it achieved the best performance.[Conclusions]First,machine learning–based ultrasound radiomics features are able to non-invasively distinguish primary liver tumors from metastatic liver tumors.Second,the radiomics signatures could non-invasively explore the underlying association between contrast enhanced CT images and pathological grades of HCC.Third,the contrast enhanced CT-based radiomics could non-invasively explore the underlying association between contrast enhanced CT images and early recurrence of HCC via machine learning radiomcis.[Innovation and Deficiency]The innovation of the present study:First,ultrasound-based radiomics was used for preoperative classification of primary versus metastatic liver cancer,and multiple machine learning-based algorithms with crossvalidation strategy were applied to consturct machine learning–based ultrasound radiomics model.Second,contrast enhanced CT-based radiomics was used for preoperative prediction of pathological grades of hepatocellular carcinoma(HCC)via machine learning.We employed contrast enhanced CT-based radiomics signatures and XGBoost classifier for preoperative prediction of pathological grades of HCC,XGBoost classifier with 5-fold cross-validation strategy was utilized to train the classification model.It could achieve beteer compared to the traditional methods.Third,contrast enhanced CT-based radiomics was used for preoperative prediction of early recurrence of hepatocellular carcinoma(HCC)via machine learning,and it achieved better performance compared to traditional methods.The deficiency of the present study:First,the experiment lacks external validation since it is a single-center cohort study.Second,The modal of the images in our study included only CT or ultrasound.