0917
Intertumoral Heterogeneity based on MRI Radiomics Features Predicts prognosis in HCC patients before Hepatectomy1radiology, Third Affiliated Hospital of Sun Yat-Sen University, Guangzhou, China, 2radiology, Meizhou People's Hospital, Meizhou, China, 3radiology, Third affliated hospital of San Yet-Sun university, Guangzhou, China
Synopsis
Keywords: Liver, Liver, hepatocellular carcinoma
Motivation: Hepatocellular carcinoma (HCC) exhibits significant intertumoral heterogeneity, which contributes significantly to treatment resistance and failure. Noninvasive imaging and radiomics for preoperative decoding of the subtypes and prognosis may be valuable in clinical management.
Goal(s): To preoperatively develop and validate clustering analysis of HCC based on MRI radiomics features for identifying subtypes with discrete prognosis.
Approach: We performed clustering analysis of HCC based on MRI radiomics features to detect distinct subtypes, and subsequently clinicopathological parameters and prognosis were compared and evaluated between different subtypes.
Results: Based on the radiomics features of MRI, clustering analysis identified two distinct subtypes with discrete prognosis in HCC patients.
Impact: Clustering analysis based on the radiomics features of multiparametric MRI is a potential noninvasive decision-making method for the management of patients with HCC in clinical practice.
Introduction
Hepatocellular carcinoma (HCC) is the fifth most common cancer and the fourth leading cause of cancer-related deaths worldwide, and with high heterogeneous and different prognosis (1). Precise diagnosis of tumor phenotypes and recurrence risk is of vital importance in the clinical management of HCC (2-4). Although imaging modalities such as CT and contrast-enhanced MRI have played an essential role in the noninvasive diagnosis and prognosis of HCC, radiomics has also shown great potential in the precise diagnosis of HCC and preoperative prediction of the recurrence risk (5-7). Radiomics-empowered image interpretation used to amplify the differences in tumor heterogeneity between different phenotypes is limited in HCC patients. In this study, we aim to use clustering analysis based on the radiomics features of contrast-enhanced MRI for identifying subtypes with discrete prognosis in HCC patients before hepatectomy.Methods
This multicenter study respectively evaluated three datasets from three independent centers (center 1, training and internal test cohort; centers 2 and 3, external test cohort) of contrast-enhanced MRI images in 450 HCC patients with histopathologic-proven from November 2016 to February 2021. Radiomics features analysis was performed on T1-weighted pre-contrast phase, late arterial phase (LAP) and portal-venous phase (PVP) of preoperative MRI scans. We implemented a three-stage filtering strategy to select reliable radiomic features. First, features with interclass correlation coefficients (ICC) below 0.9 were eliminated. Second, to capture significant variability between tumors, only those features with the highest variance (top 1) were selected. Finally, a thorough assessment of pairwise correlations was performed using Pearson's correlation coefficient, and a single random feature from each pair with a correlation value greater than 0.9 was discarded to mitigate multicollinearity concerns. Based on the selected radiomics features and a non-negative matrix factorization (NMF) approach, HCCs were classified into different subtypes in training cohort, then which were validated in internal and external test cohorts. Clinical parameters and microvascular invasion (MVI) status among distinct subtypes were compared using the Mann-Whitney U test, Kaplan-Meier survival curves and log-rank tests were used to evaluate the differences in prognosis among identified distinct subtypes in three cohorts. The multivariable Cox regression was used to evaluated the prognostic value of the identified subtypes in HCC patients.Results
A total of 450 patients were respectively included (training cohort, n = 213; internal test cohort, n = 140; external test cohort, n = 97). Patients in three cohorts had similar baseline characteristics (all P > 0.05). The optimal number of clusters in our dataset was determined by experimenting with varying numbers of clusters from 2 to 10 using the elbow method. Two image subtypes were identified in training cohort (subtype 1, n = 86; subtype 2, n =127) and also validated in internal test cohort (subtype 1, n = 60; subtype 2, n = 80) and external test cohort (subtype 1, n = 39; subtype 2, n = 58) (Fig 1). Imaging subtype 1 presented greater pre-contrast_wavelet_HHL_glcm_Idn, LAP_log_sigma_5.0mm_3D_glcm_InverseVarianc, LAP_wavelet_LLH_glszm_ZoneEntropy, LAP_wavelet_LLH_glszm_ZoneEntropy, LAP_wavelet_LHH_glrlm_GrayLevelNonUniformityNormalized and LAP_wavelet_LLL_glcm_Imc1 than subtype 2. Subtype 2 presented greater pre-contrast_log_sigma_4.0mm_3D_glrlm_RunPercentage, pre-contrast_wavelet_LHL_glcm_Imc2, pre-contrast_wavelet_HLL_glcm_Imc2, pre-contrast_wavelet_HHH_gldm_LowGrayLevelEmphasis, LAP_log_sigma_2.0mm_3D_glcm_Imc2, LAP_wavelet_HLL_glcm_Imc2, LAP_wavelet_HHL_glcm_Imc2, PVP_wavelet_LHL_glcm_Imc2 and PVP_wavelet_HLL_glcm_Imc2 than subtype1. Compared to subtype 2, HCC patients with subtype 1 showed higher AST, PLT, ALP, GGT, LDH, MVI-positive and poor recurrence-free survival among three cohorts (all P < 0.05, training cohort Fig 1D, Fig 2A; internal test cohort Fig 1D, Fig 2B and external test cohort Fig 1F, Fig 2C). At multivariable analysis, the image subtype was an independent predictor of recurrence-free survival in training cohort ((P = 0.001) (Table 1), and also validated in internal test cohort (P = 0.003) and external test cohort (P < 0.001) (Table 2). Two examples were shown in Fig 3.Discussion
In this study, our results showed that two distinct subtypes were identify by clustering analysis based on radiomics features extracted from preoperative contrast-enhanced MRI in HCC patients. There are significantly different clinical biomarkers, MVI status and outcome. The image subtype was an independent prognostic factor in HCC patients after surgical resection, which maybe a potential noninvasive decision-making method for HCC patients. Future studies are required to validate our findings.Conclusion
Clustering analysis based on MRI radiomics features can identify two image subtypes with distinct clinicopathological characteristics and prognosis in HCC patients. The image subtype was an independent predictor of recurrence-free survival.Acknowledgements
National Natural Science Foundation of China grant (82271973 and 91959118, Jin Wang), The ‘Five Five’ Project of the Third Affiliated Hospital of Sun Yat-sen University (2023WW103, Jin Wang), Guangdong Basic and Applied Research Foundation (2021A1515010582, Jin Wang), Key Research and Development Program of Guangdong Province (2019B020235002, Jin Wang), China International Medical Foundation SKY Research Fund for Medical Imaging (Z-2014-07-2101 and Z-2014-07-1912-15, Jin Wang), Clinical Research Foundation of the 3rd Affiliated Hospital of Sun Yat-Sen University (YHJH201901, Jin Wang), National Natural Science Foundation of China grant number (82202129, Chao Li)References
1. Llovet JM, Zucman-Rossi J, Pikarsky E, Sangro B, Schwartz M, Sherman M, Gores G. Hepatocellular carcinoma. Nat Rev Dis Primers. 2016 Apr 14;2:16018.
2. Kurebayashi Y, Matsuda K, Ueno A, Tsujikawa H, Yamazaki K, Masugi Y, Kwa WT, Effendi K, Hasegawa Y, Yagi H, Abe Y, Kitago M, Ojima H, Sakamoto M. Immunovascular classification of HCC reflects reciprocal interaction between immune and angiogenic tumor microenvironments. Hepatology. 2022 May;75(5):1139-1153.
3. Shimada S, Mogushi K, Akiyama Y, Furuyama T, Watanabe S, Ogura T, Ogawa K, Ono H, Mitsunori Y, Ban D, Kudo A, Arii S, Tanabe M, Wands JR, Tanaka S. Comprehensive molecular and immunological characterization of hepatocellular carcinoma. EBioMedicine. 2019 Feb;40:457-470.
4. Jiang Y, Sun A, Zhao Y, Ying W, Sun H, Yang X, Xing B, Sun W, Ren L, Hu B, Li C, Zhang L, Qin G, Zhang M, Chen N, Zhang M, Huang Y, Zhou J, Zhao Y, Liu M, Zhu X, Qiu Y, Sun Y, Huang C, Yan M, Wang M, Liu W, Tian F, Xu H, Zhou J, Wu Z, Shi T, Zhu W, Qin J, Xie L, Fan J, Qian X, He F; Chinese Human Proteome Project (CNHPP) Consortium. Proteomics identifies new therapeutic targets of early-stage hepatocellular carcinoma. Nature. 2019 Mar;567(7747):257-261.
5. Segal E, Sirlin CB, Ooi C, Adler AS, Gollub J, Chen X, Chan BK, Matcuk GR, Barry CT, Chang HY, Kuo MD. Decoding global gene expression programs in liver cancer by noninvasive imaging. Nat Biotechnol. 2007 Jun;25(6):675-80.
6. Kitao A, Matsui O, Zhang Y, Ogi T, Nakada S, Sato Y, Harada K, Yoneda N, Kozaka K, Inoue D, Yoshida K, Koda W, Yamashita T, Yamashita T, Kaneko S, Kobayashi S, Gabata T. Dynamic CT and Gadoxetic Acid-enhanced MRI Characteristics of P53-mutated Hepatocellular Carcinoma. Radiology. 2023 Feb;306(2):e220531.
7. Chen J, Wu Z, Xia C, Jiang H, Liu X, Duan T, Cao L, Ye Z, Zhang Z, Ma L, Song B, Shi Y. Noninvasive prediction of HCC with progenitor phenotype based on gadoxetic acid-enhanced MRI. Eur Radiol. 2020 Feb;30(2):1232-1242.
Figures
