Pre-surgery Enhanced Magnetic Reason Image Based-radiomics In Diagnosis For Specific Subtype Of Hepatocellular Carcinoma

Background: Liver cancer is the third leading cause of cancer-related mortality worldwide,and hepatocellular carcinoma(HCC)is the main entity accounting for 90% of all cases.Unresectable HCC and advanced HCC can only accept systematic therapy.The next generation sequencing unveiled abnormally activated PI3K-Akt-m TOR signaling during hepatocarcinogenesis.Several studies found that HCC patients and HCC patients undergoing liver transplantation,those with activated m TOR signaling,can benefit from m TOR inhibitor therapy including rapamycin and rapamycin analogues.And key molecule ribosomal protein S6 kinase(RPS6K)can serve as biomarker for activated m TOR signaling.However,there is no non-invasive and accurate test method for RPS6 K expression level in HCC patients.Identification of HCC patients based on HCC subtype is of tremendous importance,it can provide reference information for clinical decision making and follow-up management,further improving prognosis of these patients.Radiomics can realize high-throughput feature mining from medical images and subsequent classifier/diagnostic model construction.It is widely applied in oncology research,including tumor subtype diagnosis,gene/protein expression prediction,chemotherapy response evaluation,molecular phenotype classification and tumor microenvironment depiction.Objective: In this research,we attempted to utilize MRI-based radiomics and machine learning algorithms to construct non-invasive and accurate diagnostic model for RPS6 K high HCC.Our research could help realize pre-treatment classification of this subtype and further provide reference information for subsequent clinical decision making and follow-up management.Methods: We retrospectively enrolled HCC patients who received curative hepatectomy in First Affiliated Hospital Zhejiang University School of Medicine between January 2018 and May 2019.Analyzed data included clinical information,tumor specimen and pre-surgery liver gadoxetic enhanced MRI data.The RPS6 K expression was determined by IHC,and enrolled HCC patients were further randomly split into training and validation cohorts at 7:3 ratio.We used MATLAB software to finish radiomics features extraction,R language to construct clinical diagnostic model,radiomic diagnostic model and predictive nomogram.The diagnostic ability of all built models was validated and compared in the validation cohort.Results: In both training and validation cohorts,the AUC of random forest-based DWI diagnostic model were 0.831 and 0.731;the AUC of artificial neural network-based T2 diagnostic model were 0.846 and 0.736;the AUC of artificial neural network-based fusion diagnostic model were 0.887 and 0.826;the AUC of the predictive nomogram were 0.917 and 0.845.Conclusion: Both MRI-Radiomics diagnostic model,including single sequence(DWI and T2)and fusion sequence,and predictive nomogram exhibited excellent diagnostic ability,and could achieve pre-treatment diagnosis of RPS6 K high HCC in a non-invasive and accurate fashion.Background: Cytokeratin 19(CK19)positive HCC is a special subtype,and patients of this subtype were vulnerable to lymph node metastasis,radio-chemotherapy resistance,and early recurrence after curative hepatectomy,transhepatic arterial chemotherapy and embolization(TACE),radiofrequency ablation(RFA)and liver transplantation,and poor prognosis.Subtype-oriented therapy,intense follow-up and early intervention could improve the prognosis for CK19 positive HCC patients.The current diagnosis for this subtype relied on IHC staining,existing radiomics-based attempts were limited by small sample or relied on conventional radiological diagnosis.Therefore,in this study we utilized radiomics methods,time-independent multi-center clinical and pre-surgery MRI data to construct pre-treatment diagnostic model for CK19 positive HCC in a non-invasive and accurate way.Methods: We retrospectively enrolled HCC patients who received curative hepatectomy in three hospitals,including First Affiliated Hospital Zhejiang University School of Medicine,Shulan Health(Hangzhou)Hospital and Wenzhou Medical University Affiliated Lishui Central Hospital,between different time periods.Analyzed data included clinical information,tumor specimen and pre-surgery liver gadoxetic enhanced MRI data.The CK19 expression was determined by IHC.We used MATLAB software to finish radiomics features extraction,R language to finish clinical diagnostic model and radiomic diagnostic model construction.The diagnostic ability of all built models was validated and compared in the two external independent validation cohorts.Results: From univariate analysis,the serum AFP expression level in CK19 positive HCC patients is significantly higher,and CK19 positive HCC patients are much easier to develop microvascular invasion.In training and external validation cohorts,the AUC of random forest T2 radiomics classifier are 0.824,0.603 and 0.604;The AUC of ANN DWI radiomics classifier are 0.899,0.697 and 0.721;The AUC of the ANN fusion radiomics classifier are 0.899,0.753 and 0.771.Conclusions: CK19 positive HCC exhibited worse tumor biology.Based on time-independent multi-center MRI data,we successfully constructed radiomics diagnostic models,including single sequence(DWI and T2)and fusion sequence,for pretreatment diagnosis of CK19 positive HCC in a non-invasive and accurate fashion.Our models can provide reference information for subtype-oriented management for CK19 positive HCC patients.Innovative points:1.In this study,we utilized radiomics methods to extract high-throughput radiomics features of RPS6 K high HCC.With machine learning algorithms applied,we successfully constructed pre-treatment diagnostic models for RPS6 K high HCC in a non-invasive and accurate way.Our results could provide reference information for clinical decision making and personalized management for this subtype.2.Based on time-independent multi-center clinical and pre-surgery liver gadoxetic enhanced MRI data,we validated the special tumor biology of CK19 positive HCC reported previously.We further constructed radiomics feature and machine learning based diagnostic models.With external validation,the radiomic model demonstrated its ability to achieve non-invasive,accurate and efficient diagnosis for CK19 positive HCC patients.