CT-based deep learning radiomics model for preoperative prediction of microvascular invasion in intrahepatic cholangiocarcinoma and its clinical value: a two-center study

doi:10.21203/rs.3.rs-3856223/v1

Download PDF

Research Article

CT-based deep learning radiomics model for preoperative prediction of microvascular invasion in intrahepatic cholangiocarcinoma and its clinical value: a two-center study

https://doi.org/10.21203/rs.3.rs-3856223/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

To develop and validate a computed tomography (CT) image-based deep learning radiomics model (DLRAD) for preoperative prediction of MVI in ICC patients, and to validate its relationship with prognosis.

Methods

A total of 165 ICC patients were recruited from two centers for retrospective study. Based on the radiomics and deep learning features of arterial phase CT images, dozens of models were constructed and compared using four machine learning methods. The incremental value of different sizes of peritumoral regions to the model was also explored. The performance of the model was evaluated using the area under the curve (AUC), calibration curve and decision curve. Kaplan-Meier curve was used to analyze the relationship between the model prediction results and prognosis.

Results

According to the radiomics features in the intratumoral and 2mm peritumoral regions and the deep learning features in the tumor, the DLRAD model constructed by the LR method showed the best discrimination ability for MVI. The AUC of the internal validation cohort was 0.86. The AUC of the external validation cohort was 0.89. In addition, the MVI predicted by the model was significantly correlated with the overall survival rate of patients (P = 0.005), which was consistent with the actual situation.

Conclusion

The DLRAD model constructed by radiomics and deep learning technology can effectively predict MVI in ICC patients. This provides clinicians with a powerful tool to help them make more accurate treatment decisions.

intrahepatic cholangiocarcinoma

microvascular invasion

deep learning

radiomics

prognosis

Intrahepatic cholangiocarcinoma (ICC) is the second most common liver malignant tumor, which originates from intrahepatic bile duct epithelial cells. Although the incidence of intrahepatic cholangiocarcinoma is much lower than that of hepatocellular carcinoma, the overall incidence of ICC has been increasing in recent decades. Surgery is the main treatment for resectable ICC^[1–3]. However, the prognosis is still not ideal. Among them, the high postoperative recurrence rate and high metastasis rate are the main reasons^[4]. Microvascular invasion (MVI) is a poor prognostic factor for ICC, which is closely related to postoperative recurrence and metastasis^{[5, 6]}.

Preoperative prediction of MVI is of great significance. Microvascular invasion, which refers to the invasion and destruction of tumor cells in the microvessels of the tumor, usually indicates an aggressive and metastatic potential. Understanding microvascular invasion allows doctors to assess the malignancy of the tumor, which in turn improves the accuracy of patient prognosis evaluation. Second, preoperative prediction of MVI can guide appropriate treatment and preventive measures to be taken. For patients with microvascular invasion, more radical surgical resection or other therapeutic measures may be recommended to reduce the risk of tumor recurrence. Furthermore, understanding MVI can also assist doctors in evaluating patient response to immunotherapy, as immunotherapy drugs may be more effective in such patients. Therefore, preoperative determination of MVI status in ICC patients is helpful for formulating treatment strategies and individualized monitoring plans^[7–10].

Currently, preoperative assessment of biologic characteristics is normally performed by obtaining tissue samples by means of needle aspiration. However, due to the limited size of the tissue specimen obtained by biopsy, the assessment of the biological characteristics of the biopsy sample may not be representative of the whole tumor. Tumor development and change is a dynamic process, so its changes in time and space are heterogeneous^[11]. Fine needle aspiration not only increases the risk of needle and blood dissemination, but also leads to complications such as post-puncture bleeding and puncture site infection^[12]. The proficiency of the operators affects the accuracy of puncture results and the risk of complications.

Radiomics has received extensive attention since it was proposed in 2012^[13]. This technology enables high-throughput selection and identification of imaging features from medical images that are not recognizable by human eyes, providing essential supplementary data to help reveal the potential relationship between quantitative image features and heterogeneity. Deep learning has made remarkable progress in medical image analysis and has rapidly promoted the development of this field. Residual nets (RESNET) is a common deep learning (DL) method that can realize various clinical tasks such as disease classification and diagnosis by analyzing medical images^[14]. For example, with a single preoperative computed tomography (CT) image examination, patients can obtain a comprehensive quantitative assessment of liver tumors efficiently and conveniently. Today, the development of radiomics in cancer diagnosis, treatment decision making, and prognosis prediction is encouraging. Many investigators are applying this approach to the field of intrahepatic cholangiocarcinoma^[15–17].

The features selected by the computer and those recognized by the human eye form complementary relationships. We have an opportunity to improve the accuracy of our medical practice by working with human-machine collaboration. In this study, we aim to develop and validate an integrated radiomics and DL feature model for preoperative prediction of MVI in ICC, which could help in treatment planning and prognosis evaluation.

Patients:

The research adhered to the principles of the Declaration of Helsinki and was approved by the ethics committee of the People's Hospital of Henan University and Henan Cancer Hospital (Ref No. 2023–012 and Ref No. 2023–203)

The clinical data of ICC patients in Henan Provincial People's Hospital and Henan Cancer Hospital were retrospectively analyzed. Inclusion criteria: 1. Intrahepatic cholangiocarcinoma was confirmed by histopathology after surgical resection, and had MVI status; 2. Preoperative enhanced CT examination within 2 weeks; 3. The CT images are clear and free of artifacts and defects. Exclusion criteria: 1. Liver metastasis or combined with other malignant tumors; 2. The patient had received anti-tumor therapy before the operation; 3. The maximum diameter of the tumor is less than 1cm, which is not easy to delineate. The flow chart of patient screening is shown in Fig. 1.

Overall survival (OS) was calculated as the interval between the date of surgery and the date of death or last follow-up. Follow-up was conducted by telephone, outpatient examination, and inpatient review until July 2022.

CT image acquisition and tumor segmentation:

The study conformed to the International Biomarker Standardization Initiative. All patients were evaluated with enhanced CT, and arterial phase CT images with a thickness of 5mm were obtained. Specific imaging parameters are detailed in Supplementary Material A. Before radiomics feature extraction, in order to improve the standardization of CT images, the images were preprocessed as follows: (1) The gray value of the image was uniformly 350/40HU; (2) The image was resampled and the voxel size was 1×1×1mm. The tumor region of interest (ROI) was manually delineated on CT images by experienced clinicians using the open source software 3D Slicer. After one week, an additional 30 patients were randomly selected and the ROIs were again segmented by an additional clinician to construct two resegmented datasets that could be used to assess the reproducibility of radiomics features.

Feature Extraction:

The work flow chart of the study is shown in Fig. 2. In the radiomics workflow, 729 radiomics features were extracted using the Pyradiomics software including wavelet_Firstorder, wavelet_GLCM, wavelet_GLDM, wavelet_GLRLM, wavelet_NGTDM and wavelet_GLSZM. This is because wavelet features can effectively capture image changes at different scales and frequencies, thus enabling accurate and detailed analysis and interpretation of images. A large number of previous studies have also shown the importance of wavelet features in model development^[18–20]. These wavelet features provide rich tumor information and contribute to a deeper understanding and interpretation of tumor characteristics. In the deep transfer learning step, the pre-trained ResNet34 model was implemented in the Pytorch framework and 512 DL features were extracted.

Feature extraction and statistical analysis:

To eliminate scaling differences between radiomics and DL features, feature normalization was performed before feature selection using the z-score method. For features with high repeatability, we used Spearman's rank correlation coefficient to calculate the correlation between features, and one feature was retained that has a correlation coefficient greater than 0.9 with any other feature. To maximize the ability to depict features, we employed a greedy recursive deletion strategy for feature filtering, which means that the feature with the greatest redundancy in the current set is deleted each time. Subsequently, we applied LASSO regression model to the discovery dataset for signature construction. Depending on the regulation weight λ, LASSO shrinks all regression coefficients toward zero, effectively eliminating irrelevant features. Through 10-fold cross validation, we were able to find the optimal λ value that yielded minimum cross validation error and retained features with nonzero coefficients for the regression model. Finally, these features were combined into a radiomics signature, and a radiomics score for each patient was obtained by linearly combining the retained features weighted by their model coefficients.

Python software (Python 3.11, Jupyter notebook environment) and SPSS (v26.0, IBM) statistical software were used for statistical analysis. Kaplan-Meier curve was used for survival analysis, and log-rank test was used for comparison between groups. LASSO analysis, ROC analysis, calibration plot, survival analysis and DCA were performed using Python statistical software. Python implements deep learning network models. Two-sided p value < 0.05 was considered statistically significant.

Model Construction:

After completing Lasso feature selection, we input the final features into Logistic regression (LR), Support vector machine (SVM), K-nearest neighbor (KNN), and Naive-Bayes (NB) machine learning models to construct prediction models. We designed the following four types of models: (1) intratumoral radiomics models (INTRA): We focused on analyzing the predictive value of intratumoral radiomics features for MVI. (2) Combined intratumoral and peritumoral models (INTRAPER): The intratumoral features and peritumoral features of different sizes were fused to explore the influence of the tumor environment on tumor progression and evaluate its contribution to the performance of the model. (3) Deep transfer learning model (DL): deep learning technology was used to improve the performance of the model. (4) Combination of deep learning and radiomics model (DLRAD): to give full play to the advantages of deep learning and radiomics to predict tumor MVI more comprehensively. We also tried to reveal the influence of tumor characteristics on disease progression. To comprehensively evaluate the performance of the model, we plotted ROC curves to measure the calibration efficiency of the model by a calibration curve, and used Hosmer-Lemeshow analysis to evaluate the adaptability and robustness of the model. Finally, decision curve analysis (DCA) was used to evaluate the practicability and generalization ability of the prediction model.

Clinical utility of model:

A large number of studies have confirmed that MVI is an independent risk factor for poor prognosis of ICC patients. According to the prediction results of the model, patients were divided into high-risk group and low-risk group to compare the OS of the two groups. This study not only aims to verify the effect of MVI on the prognosis of ICC patients, but also confirms the practical application value of this model in clinical practice.

Construction and validation of radiomics and deep learning models:

Through the LASSO regression based on the training cohort, 14 radiomics features and 10 DL features were finally extracted from the CT arterial phase images of each patient, which were then used to construct and calculate the scores of radiomics signature and DL signature, respectively. The scores are formulated by linearly combining these features with the corresponding LASSO coefficients to improve the accuracy and precision of the scores. The scores formula is shown in the Supplementary material B.

To construct the prediction model, the final features were fed into four machine learning models, including LR, SVM, KNN, and NB. The AUC/Accuracy/Sensitivity of the RAD and DL models for predicting microvascular invasion of intrahepatic cholangiocarcinoma are shown in Table 1. According to the training and test the performance of the queue, visible LR and SVM model compared to other machine learning models has better performance.

Table 1

Performance of radiomics feature model and deep transfer learning feature model.
Training cohort
Type	Model-name	AUC	95% CI	Accuracy	Sensitivity	Specificity
RAD	LR	0.888	0.8216–0.9536	0.830	0.766	0.887
	NB	0.797	0.7104–0.8834	0.750	0.574	0.906
	SVM	0.928	0.8726–0.9833	0.890	0.851	0.925
	KNN	0.850	0.7783–0.9218	0.670	0.340	0.962
DL	LR	0.819	0.7363–0.9016	0.760	0.723	0.792
	NB	0.872	0.8022–0.9424	0.810	0.745	0.868
	SVM	0.956	0.9112–1.0000	0.910	0.872	0.943
	KNN	0.833	0.7534–0.9126	0.640	0.277	0.962
Internal validation cohort
Type	Model-name	AUC	95% CI	Accuracy	Sensitivity	Specificity
RAD	LR	0.706	0.5450–0.8673	0.674	0.632	0.708
	NB	0.647	0.4811–0.8127	0.605	0.842	0.417
	SVM	0.640	0.4633–0.8174	0.674	0.368	0.917
	KNN	0.614	0.4461–0.7820	0.605	0.421	0.750
DL	LR	0.741	0.5932–0.8893	0.674	0.789	0.583
	NB	0.748	0.6000–0.8956	0.651	0.789	0.542
	SVM	0.730	0.5769–0.8836	0.674	0.684	0.667
	KNN	0.692	0.5384–0.8453	0.605	0.632	0.583
External validation cohort
Type	Model-name	AUC	95% CI	Accuracy	Sensitivity	Specificity
RAD	LR	0.726	0.5020–0.9510	0.727	0.667	0.769
	NB	0.650	0.3959–0.9033	0.682	0.333	0.923
	SVM	0.726	0.5023–0.9507	0.682	0.778	0.615
	KNN	0.470	0.1954–0.7448	0.545	0.000	0.923
DL	LR	0.812	0.6304–0.9935	0.727	0.667	0.769
	NB	0.769	0.5206–1.0000	0.818	0.778	0.846
	SVM	0.855	0.6878–1.0000	0.773	0.778	0.769
	KNN	0.752	0.5271–0.9772	0.636	0.333	0.846
AUC Area Under Curve, CI confidence interval, RAD radiomics features, DL deep learning features, LR logistic regression, NB Naive-Bayes, SVM Support vector machine, KNN K-nearest neighbor.

Incremental value of the peritumoral region to the model:

Peritumoral tissue has been shown to provide useful information for diagnosis and prediction of prognosis. The main objective of this study was to evaluate its diagnostic performance by using peritumoral areas of three sizes. For the prediction of MVI, the LR and SVM models constructed using the radiomics features of the intratumoral region and the 2mm peritumoral region showed better performance, as shown in Fig. 3. Among the 22 radiomics features used for modeling, 7 features were intratumoral (LLL_ngtdm_Busyness, HLH_glszm_LargeAreaLowGrayLevelEmphasis, LLL_firstorder_Skewness, LHL_firstorder_Median, HLH_firstorder_Median, LLH_glcm_ClusterShade and LLL_firstorder_Kurtosis) and 15 features were peritumoral (HLL_glcm_Correlation, LHL_glcm_Correlation, LLL_firstorder_Skewness, LHH_firstorder_Median, HLL_glcm_lmc2, HLL_firstorder_90Percentile, HHH_glcm_ClusterProminence, HLH_glcm_ClusterShade, HHL_firstorder_Mean, HLH_ngtdm_Busyness, LLL_glcm_Correlation, HLL_firstorder_10Percentile, LHL_firstorder_Median, LHL_firstorder_Mean and LLL_ngtdm_Strength). In the training cohort, the AUC of LR model and SVM model were 0.942 and 0.963, respectively. In the internal validation cohort, the AUC of LR model and SVM model was 0.792 and 0.754, respectively. External validation queue, the AUC 0.795 LR model, the AUC of the SVM model is 0.94. Supplementary Material C summarizes the use within the tumor area and around 1 mm, 2 mm, 3 mm tumor area to build the performance of the model.

Model fusion and clinical application value:

The final model (DLRAD) was constructed by combining the radiomics features and DL features of the intratumoral region with the radiomics features of the 2mm peritumoral region according to the above feature dimensionality reduction algorithm, which showed the best performance. The AUC of the internal validation cohort was 0.86 (0.7455–0.9738), the accuracy was 0.814, the sensitivity was 0.842, and the specificity was 0.792. The AUC of the external validation cohort was 0.889 (0.7487-1.0000), the accuracy was 0.818, the sensitivity was 0.889, and the specificity was 0.769 (Fig. 4a). Calibration curve of the model shows that in the three cohorts, forecast and has an excellent consistency between the observed MVI (Fig. 4b). The Hosmer-Lemeshow test showed P = 0.198 for the internal validation cohort and P = 0.171 for the external validation cohort, showing the applicability of the DLRAD model in the two validation cohorts. Decision-making curve shows that the model has an obvious positive net income, has great clinical practical (Fig. 4c).

As of July 2022, 83/163 patients (50.92%) had completed OS follow-up. To analyze the clinical value of the model, patients were divided into high-risk and low-risk MVI groups based on the predictions of the model. Survival curves showed (Fig. 5) that in the follow-up cohort, the median OS was 25.7 months (95%CI 27.8–29.6) in the low-risk group and 17.3 months (95%CI 12.5–22.1) in the high-risk group (P = 0.005).

In this study, we finally constructed the DLRAD model for preoperative prediction of MVI status in ICC patients. The proposed model combined radiomics features from the intratumoral region and the 2mm peritumoral region, and introduced deep transfer learning features based on RENET34, which showed satisfactory prediction performance. Based on the MVI results predicted by this model, we can judge the prognosis of patients, and this finding has important implications for clinical decision making because it can help physicians to more accurately assess the risk of patients and thus formulate individualized treatment plans.

In the process of model building, different machine learning methods are suitable for different tasks, and choosing appropriate algorithms can improve the accuracy and efficiency of the analysis. The model was initially trained and validated by applying different machine learning algorithms for model building, including LR, SVM, KNN and NB. The results showed that the LR and SVM methods were more suitable for our task and better able to analyze the radiomics data, which was consistent with the results of previous studies^{[21, 22]}.

In addition to the tumor itself, the peritumoral area is also an important predictor of MVI. Fiz et al^[23]. performed non-invasive identification of MVI in intrahepatic cholangiocarcinoma by preoperative PET/CT radiomics features. The results showed that the inclusion of peritumoral features further improved the prediction performance. However, this study did not compare peritumoral features of different sizes and was based on a single-center study of 74 patients. In our study, we not only considered the imaging features inside the tumor, but also focused on the characteristics of the peritumoral region. Through comparative analysis, we found that the features of the peritumoral region had obvious incremental value for the prediction accuracy of the model. This further demonstrates the value of radiomics in the preoperative evaluation of tumors and provides more comprehensive information for clinical practice. In addition, the internal validation cohort model results suggest that the appropriate regional tumor weeks has contribution value, the construction of a model according to the tumor and tumor area 1 mm, 2 mm, 3 mm build the model of AUC LR respectively (0.757, 0.792, 0.645), and the SVM (0.732, 0.754, 0.627); in the external validation cohort LR, (0.744, 0.795, 0.897) and SVM (0.778, 0.94, 0.897) were used. Because the number of external validation cohorts was too limited, we could not conclude that the incremental value of a 2mm peritumoral region on model performance was higher, although this pattern was observed in the internal validation cohorts.

Deep learning techniques have shown great potential for tumor diagnosis and prediction^{[24, 25]}. By analyzing a large amount of tumor image data, deep learning models can extract meaningful features from images, and then predict the pathological characteristics of tumors. Since there are relatively few ICC patients and renet34 transfer learning is more suitable for tasks with relatively limited samples, we used this model to predict microvascular invasion in this study and achieved satisfactory accuracy.

This study has the following advantages over previous studies^{[23, 26, 27]}. First, we explored the value of different machine learning methods for MVI prediction. Secondly, the application of deep transfer learning algorithm improved the prediction accuracy of the model and further enriched the research methods of radiomics. In addition, we extracted features from both intratumoral and peritumoral regions, which more comprehensively reflected the characteristics of the tumor. Nonetheless, this study has some limitations. In future research, we can further explore more advanced deep learning models and optimization algorithms to improve the prediction accuracy and generalization ability of the model. At the same time, we need to expand the sample size to improve the reliability and generalizability of the study, and "virtual biopsy" is our goal.

In summary, this study will attempt to explore the potential ability of CT in preoperative prediction with the aim of developing a feasible and robust model. The application of deep learning and radiomics methods in predicting microvascular invasion of intrahepatic cholangiocarcinoma has achieved great results.

Intrahepatic cholangiocarcinoma (ICC);

Microvascular invasion (MVI);

Deep learning (DL);

Residual nets (RESNET);

Computed tomography (CT);

Overall survival (OS);

Region of interest (ROI);

Logistic regression (LR);

Support vector machine (SVM);

K-nearest neighbor (KNN);

Naive-Bayes (NB);

Intratumoral radiomics models (INTRA);

Combined intratumoral and peritumoral models (INTRAPER);

Combination of deep learning and radiomics model (DLRAD);

Area Under Curve (AUC);

Receiver operating characteristic curve (ROC).

Ethics approval and consent to participate:

Consent for publication

All authors consented to the submission of this manuscript.

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

The study was funded by Science and Technique Project of Henan province (Grant NO. 222102310709), Henan Young and Middle aged Health Science and Technology Innovation Leading Talents Training Project (Grant NO. YXKC2022002).

Authors' contributions

Conception of study: PC, ZY, PN and HY. Collecting literature: PC. Figures and tables: PC. All authors contributed to writing of the manuscript and approved the final manuscript. All authors contributed to the article and approved the submitted version.

Acknowledgements

Not applicable.

RIZVI S, KHAN S A, HALLEMEIER C L, et al. Cholangiocarcinoma — evolving concepts and therapeutic strategies. Nature Reviews Clinical Oncology, 2017, 15(2): 95–111.
SAPISOCHIN G, IVANICS T, SUBRAMANIAN V, et al. Multidisciplinary treatment for hilar and intrahepatic cholangiocarcinoma: A review of the general principles. International Journal of Surgery, 2020, 82: 77–81.
BRIDGEWATER J, GALLE P R, KHAN S A, et al. Guidelines for the diagnosis and management of intrahepatic cholangiocarcinoma. J Hepatol, 2014, 60(6): 1268–89.
KUBO S, SHINKAWA H, ASAOKA Y, et al. Liver Cancer Study Group of Japan Clinical Practice Guidelines for Intrahepatic Cholangiocarcinoma. Liver Cancer, 2022, 11(4): 290–314.
WANG Y, LI J, XIA Y, et al. Prognostic nomogram for intrahepatic cholangiocarcinoma after partial hepatectomy. J Clin Oncol, 2013, 31(9): 1188–95.
ZHANG H, YANG T, WU M, et al. Intrahepatic cholangiocarcinoma: Epidemiology, risk factors, diagnosis and surgical management. Cancer letters, 2016, 379(2): 198–205.
CREMEN S, KELLY M E, GALLAGHER T K. The role of neo-adjuvant therapy in cholangiocarcinoma: A systematic review. Frontiers in oncology, 2022, 12.
LE ROY B, GELLI M, PITTAU G, et al. Neoadjuvant chemotherapy for initially unresectable intrahepatic cholangiocarcinoma. British Journal of Surgery, 2018, 105(7): 839–47.
DHANASEKARAN R, HEMMING A W, ZENDEJAS I, et al. Treatment outcomes and prognostic factors of intrahepatic cholangiocarcinoma. Oncology reports, 2013, 29(4): 1259–67.
MORIS D, PALTA M, KIM C, et al. Advances in the treatment of intrahepatic cholangiocarcinoma: An overview of the current and future therapeutic landscape for clinicians. CA: A Cancer Journal for Clinicians, 2022, 73(2): 198–222.
BIZZARRI M, FEDELI V, MONTI N, et al. Personalization of medical treatments in oncology: time for rethinking the disease concept to improve individual outcomes. EPMA Journal, 2021, 12(4): 545–58.
SILVA M A, HEGAB B, HYDE C, et al. Needle track seeding following biopsy of liver lesions in the diagnosis of hepatocellular cancer: a systematic review and meta-analysis. Gut, 2008, 57(11): 1592–6.
LAMBIN P, RIOS-VELAZQUEZ E, LEIJENAAR R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. European journal of cancer (Oxford, England: 1990), 2012, 48(4): 441–6.
HE K, ZHANG X, REN S, et al. Deep Residual Learning for Image Recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016: 770 – 8.10.1109/cvpr.2016.90
CHEN P, YANG Z, ZHANG H, et al. Personalized intrahepatic cholangiocarcinoma prognosis prediction using radiomics: Application and development trend. Frontiers in oncology, 2023, 13.
ZHANG J, WU Z, ZHANG X, et al. Machine learning: an approach to preoperatively predict PD-1/PD-L1 expression and outcome in intrahepatic cholangiocarcinoma using MRI biomarkers. ESMO open, 2020, 5(6): e000910.
CHU H, LIU Z, LIANG W, et al. Radiomics using CT images for preoperative prediction of futile resection in intrahepatic cholangiocarcinoma. European radiology, 2021, 31(4): 2368–76.
CHETAN M R, GLEESON F V. Radiomics in predicting treatment response in non-small-cell lung cancer: current status, challenges and future perspectives. European radiology, 2020, 31(2): 1049–58.
LI M-D, CHENG M-Q, CHEN L-D, et al. Reproducibility of radiomics features from ultrasound images: influence of image acquisition and processing. European radiology, 2022, 32(9): 5843–51.
TANG V H, DUONG S T M, NGUYEN C D T, et al. Wavelet radiomics features from multiphase CT images for screening hepatocellular carcinoma: analysis and comparison. Scientific reports, 2023, 13(1).
ZHOU Y, ZHOU G, ZHANG J, et al. Radiomics signature on dynamic contrast-enhanced MR images: a potential imaging biomarker for prediction of microvascular invasion in mass-forming intrahepatic cholangiocarcinoma. European radiology, 2021, 31(9): 6846–55.
ZHANG S, HUANG S, HE W, et al. Radiomics-Based Preoperative Prediction of Lymph Node Metastasis in Intrahepatic Cholangiocarcinoma Using Contrast-Enhanced Computed Tomography. Ann Surg Oncol, 2022, 29(11): 6786–99.
FIZ F, MASCI C, COSTA G, et al. PET/CT-based radiomics of mass-forming intrahepatic cholangiocarcinoma improves prediction of pathology data and survival. European journal of nuclear medicine and molecular imaging, 2022, 49(10): 3387–400.
DONG D, FANG M J, TANG L, et al. Deep learning radiomic nomogram can predict the number of lymph node metastasis in locally advanced gastric cancer: an international multicenter study. Ann Oncol, 2020, 31(7): 912–20.
ZHONG L, DONG D, FANG X, et al. A deep learning-based radiomic nomogram for prognosis and treatment decision in advanced nasopharyngeal carcinoma: A multicentre study. EBioMedicine, 2021, 70.
PENG Y T, ZHOU C Y, LIN P, et al. Preoperative Ultrasound Radiomics Signatures for Noninvasive Evaluation of Biological Characteristics of Intrahepatic Cholangiocarcinoma. Academic radiology, 2020, 27(6): 785–97.
JIANG C, ZHAO L, XIN B, et al. (18)F-FDG PET/CT radiomic analysis for classifying and predicting microvascular invasion in hepatocellular carcinoma and intrahepatic cholangiocarcinoma. Quantitative imaging in medicine and surgery, 2022, 12(8): 4135–50.

No competing interests reported.

Supplementarymaterial.docx

Download PDF

Editorial decision: Revision requested
23 Apr, 2024
Reviews received at journal
16 Apr, 2024
Reviewers agreed at journal
07 Apr, 2024
Reviews received at journal
18 Mar, 2024
Reviewers agreed at journal
17 Mar, 2024
Reviewers invited by journal
11 Mar, 2024
Editor assigned by journal
20 Jan, 2024
Submission checks completed at journal
13 Jan, 2024
First submitted to journal
12 Jan, 2024

You are reading this latest preprint version

CT-based deep learning radiomics model for preoperative prediction of microvascular invasion in intrahepatic cholangiocarcinoma and its clinical value: a two-center study

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Background

Methods

Patients:

CT image acquisition and tumor segmentation:

Feature Extraction:

Feature extraction and statistical analysis:

Model Construction:

Clinical utility of model:

Results

Construction and validation of radiomics and deep learning models:

Incremental value of the peritumoral region to the model:

Discussion

Conclusions

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1