Early prediction of local tumor progression after ablation of colorectal liver metastases based on MRI radiomics

doi:10.21203/rs.3.rs-4895617/v1

Download PDF

Article

Early prediction of local tumor progression after ablation of colorectal liver metastases based on MRI radiomics

https://doi.org/10.21203/rs.3.rs-4895617/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

The aim of this study was to investigate a magnetic resonance imaging(MRI)-based model for early prediction of local tumor progression (LTP) after ablation of colorectal cancer liver metastases (CRLM).53 patients with CRLM ablation were enrolled in a total of 83 lesions. The lesions were divided into LTP group (n = 27) and non-LTP group (n = 56). Radiomics features were extracted from the patients' post ablation enhanced MRI arterial phase in the ablation zone, and around the ablation zone (0-5mm, 0-10mm, 5-15mm) to establish radiomics, clinical and combined models. Tumor size correlated with high LTP after ablation (p < 0.05). The 0–10 mm radiomics model around the ablation zone showed good performance, with an area under the curve (AUC) of 0.874 for the training set and 0.831 for the validation set. In addition, the diagnostic efficacy of the combined model of PAZ2 and AZ as well as the combined model of AZ-PAZ2 and clinical risk factors was superior, with an AUC of 0.896 and 0.919, respectively, for the training set, and AUC of 0.882 and 0.875, respectively, for the validation set. were 0.882 and 0.875 for the validation group, respectively. In conclusion, the combined postoperative ablation zone and peri-ablation zone radiomics model can early predict LTP after ablation in CRLM patients.

MRI

Radiomics

Colorectal liver metastasis

Ablation

Local tumor progression

Colorectal cancer is the most common gastrointestinal tract cancer, accounting for 10.0% of all malignancies diagnosed and being the second leading cause of cancer-related deaths with a 9.4% mortality rate¹. A staggering 20% of newly diagnosed colorectal cancer patients are found to have distant metastases at the time of diagnosis². The occurrence of distant metastases reduces the five-year overall survival (OS) of colorectal cancer patients from 70–18%³. The liver is often the preferred site of metastasis due to its anatomical proximity and its circulation through the portal venous system. For patients with colorectal liver metastases (CRLM), surgical resection of the primary tumor and liver metastases combined with chemotherapy is the optimal treatment, which can improve OS to 50–60%^4–6, but only 20% of patients are suitable for resection at the time of diagnosis⁷.

Radiofrequency ablation (RFA) or microwave ablation (MWA) have been recommended as a suitable treatment for patients with early-stage CRLM who are not candidates for surgical resection, especially for patients with CRLM ≤ 3 cm in diameter. For these patients, ablation may be prioritized over hepatic resection because of its comparable survival benefit, less invasive nature, and cost-effectiveness^8–12. The rate of local tumor progression (LTP) (defined as the appearance of a new tumor within 10 mm of the tumor periphery on imaging) is a major challenge in CRLM patients after thermal ablation, and has been reported to occur in 6–46% of cases^13–15. MRI has been shown to perform significantly better than CE-CT and PET/CT for the diagnosis of CRLM and detection after ablation therapy^16–18. However, MRI has a low sensitivity for detecting LTP in the first follow-up scan after ablation, and serial scans are usually required to improve accuracy. Therefore, accurate assessment of treatment response and patient prognosis is essential to optimize individualized management strategies in the increasing use of thermal ablation in CRLM.

Radiomics is defined as the high-throughput mining and quantification of imaging images, which can provide essential information about the tumor and the microenvironment surrounding the tumor^19,20. It has been influential in the direction of oncology in the prediction of recurrence, prognosis, survival, and differential diagnosis^21–23. Shahveranova et al ²⁴ explored the importance of T1 fat suppression in early arterial phases based on the T2 fat suppression and the MRI radiomics features are important predictors of LTP after MWA in patients with CRLM. Shang et al ²⁵ Radiomics extracted from the peri-tumor region could add additional value in predicting EGFR mutation status in lung adenocarcinoma patients. Radiomics has been shown to extract important biological information from the area around the tumor, which may be a potential prognostic biomarker. Therefore, radiomics analysis has become an essential tool in precision medicine²⁶. In this study, we investigated postoperative MRI radiomics based on the AZ, peri-ablation zone (PAZ), and combined AZ and PAZ for early LTP prediction in CRLM patients. This may provide a new basis for personalized medicine and close follow-up or complementary therapy for high-risk patients.

Patients

The hospital ethics committee has approved this retrospective study, waiving the requirement for informed consent. All methods were performed in accordance with the relevant guidelines and regulations. Patients with CRLM who underwent thermal ablation procedures (RFA or MWA) from January 2016 to August 2023 were included in the study. Inclusion criteria were (1) primary tumor histopathologically confirmed as colorectal cancer, (2) availability of enhanced MRI images of adequate quality within 3 months after ablation, (3) > 6 months of follow-up, and (4) complete clinical data. Exclusion criteria were (1) number of metastases > 5; (2) metastases > 3 cm in diameter; (3) hepatic treatments such as transhepatic arterial chemoembolization (TACE), radioactive particle implantation, which may affect the entire liver; (4) a second ablation (i.e., reablation) of CRLM; and (5) multiple ablation areas that were indistinguishable due to proximity resulting in fusion. A total of 53 patients with a total of 83 lesions were finally included(Fig. 1). All patients after thermal ablation were treated with neoadjuvant and adjuvant therapy consisting of standard regimens of XELOX (capecitabine and oxaliplatin), FOLFOX (fluorouracil, folinic acid, and oxaliplatin), or FOLFIRI (fluorouracil, folinic acid, and irinotecan). The date of ablation was recorded for all patients along with the date of LTP or the date of last follow-up without progression. LTP was defined based on imaging findings of abnormal nodular, diffuse, and/or abnormal peripheral enhancement around the ablation site in treated patients. Follow-up after thermal ablation consists of liver imaging (usually CE-CT or MRI) and laboratory tests 4–6 weeks after ablation and every 3 months for the first year. semiannually for the next 2 years, and then annually until 5 years postprocedure.

Ablation Procedure

Ablation procedures were conducted in the CT room under CT guidance or combined ultrasound guidance and in the operating room under ultrasound guidance, following the manufacturer's protocols. The choice of RFA or MWA was based on preference and availability. The technical success of the ablation was determined by the interventional radiologist during the procedure, observing areas of complete hypodense ablation with adequate visual margins on CT images. For ultrasound-guided ablation in the operating room, an appropriate hypodense ring around the ablated lesion was used as a therapeutic endpoint. Postoperative care included keeping patients lying down for 3 hours and closely monitoring for any adverse effects.

Radiomics Analysis

The flowchart of the study is shown in Fig. 2.

Image Acquisition

The arterial-phase images of dynamic-enhanced MRI of the abdomen were collected from each patient within 3 months after ablation. All MRI examinations were performed using two MRI scanners (Germany, Siemens 1.5T and Siemens 3.0T). The main sequences and scanning parameters of the plain axial plane: field of view 380mm×380mm, layer thickness 6mm, spacing 1.2mm; the main parameters of the 1.5T: T2WI lipid-suppression sequences (TR2200ms, TE105ms), T1WI lipid-suppression sequences (TR7.07ms, TE2.39ms), DWI sequences (with b-values taken as 0 and 800s/mm2, TR5100ms, TE2.39ms, and DWI sequences (b-value taken as 0 and 800s/mm2, TR5100ms, TE2.39ms, and DWI sequences), and DWI sequences (with b-values taken as 0 and 800s/mm2, TR5100ms, and DWI sequences) were performed. 3.0T main parameters: T2WI lipoinhibitory sequence (TR1200ms, TE81ms), T1WI lipoinhibitory sequence (TR3.25ms, TE1.13ms), DWI sequence (b-value taken as 0 and 800s/mm2, TR7400ms, TE73ms). Gadolinium diamine (Onyx) was used at a dose of 0.1 mmol/kg and an injection flow rate of 3 mL/s for dynamic enhancement imaging in the arterial phase (18s delay), portal venous phase (50s delay), and delayed phase (120s delay) after the injection of the contrast agent.

Image Segmentation

To extract AZ radiomic features, AZ segmentation was performed by a radiologist with more than 10 years of experience in diagnostic abdominal imaging. For each patient, a three-dimensional volume of interest (VOI) of the AZ was manually segmented in the arterial phase using ITK-SNAP software (version 3.8.0, http://www.itksnap.org). The radiologist was unaware of the purpose of the study. Another interventional radiologist with 20 years of experience confirmed the segmentation results. To automatically expand the area around AZ, 3D Slicer software was used with dimensions: 0-5mm, 0-10mm, and 5-15mm, and then the corresponding VOIs were synthesized. After automatic expansion, the portion outside the liver parenchyma was manually removed. The resulting VOI was then used to extract radiomic features for further analysis.

Feature Extraction

The segmented images were imported into Darwin scientific research software (https://arxiv.org/abs/2009.00908) for radiomics feature extraction. Each region of interest (ROI) contains 1781 features, including first-order features, shape-based 3D features, and texture features. Texture features include gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level size zones (GLSZM), neighboring gray level difference matrix (NGTDM), and gray level difference method (GLDM). Filtering was applied to the original image using 8 filters to generate 8 filtered transformed images. These filters include Laplace Gaussian (LoG), wavelet, square, square root, logarithmic, exponential, gradient, and local binary pattern (LBP). All extracted radiological features were normalized using Z-score. More information about the above image features can be found at https://pyradiomics.readthedocs.io/en/latest/features.html.

Radiomic feature selection and model building

The dataset was randomly divided into a training set (70%) and a validation set (30%). The optimal features were selected using the Least Absolute Shrinkage and Selection Operator Variable Selection (LASSO) method and validated by tenfold cross validation, thus reducing the possibility of overfitting. Based on their weighting coefficients and radiomics labels, Radscore was calculated for each patient. Radiomics models were built for the ablation zone (AZ), 0–5 mm around the ablation zone (PAZ1), 0–10 mm around the ablation zone (PAZ2), and 5–15 mm around the ablation zone (PAZ3) regions. The eigenvalues with weighting coefficients greater than 0.4 in the AZ and PAZ models were selected to construct the AZ-PAZ combined model. Finally, the AZ-PAZ combined model was combined with clinical risk factors to construct the combined model (C).

Statistical analysis

SPSS (version 25.0, USA) was used for statistical analysis of data. Shapiro-Wilk (S-W) test was performed for normally distributed values, and Mann-Whitney U test was performed for non-normally distributed values for measurement data. Count data were expressed as percentages, and comparisons between groups were made using methods such as the chi-square test or Fisher's exact test. Radiomics scores and classification performance of the models were assessed by receiver operating characteristic (ROC) curve analysis. Performance criteria such as area under the curve (AUC), sensitivity, and specificity were obtained. The significant differences between AUCs were assessed using the Delong test, which was used to determine the statistical significance between different models. P-values < 0.05 were considered statistically significant.

Clinical Characteristics

The study included a total of 53 patients, including 83 tumors undergoing thermal ablation. Among the 83 CRLM lesions, 27 (33%) developed LTP after ablation at a median time of 10 months (range 1–24 months). The average age of all patients was 61 ± 24 years old. The median time between ablation and post-ablation review MR interval was 57 days (range 4–89 days). The median time to progression-free survival (PFS) was 15 months (range 1–72 months). Univariate analysis identified tumor size as an independent predictor of LTP development, and there were no differences in age, gender, intrahepatic location, or ablation modality between patients with or without LTP. These clinical characteristics are summarized in Table 1.

Table 1

Clinical characteristics between Non-LTP group and LTP group
Parameters	Non-LTP group (n = 56)	LTP group (n = 27)	P
Age(years)	64(range44-83)	67(range37-85)	0.870
Gender			0.713
Male	35(62.5%)	18(66.7%)
Female	21(37.5%)	9(33.3%)
Liver lobe			0.629
Right	43(76.8%)	22(81.5%)
Left	13(23.2%)	5(18.5%)
Tumor size(mm)	10(range5-24)	19(7–30)	<0.001*
Ablation			0.149
RFA	53(94.6%)	23(85.2%)
MWA	3(5.4%)	4(14.8%)

Radiomics Features

A total of 1781 radiomics features were extracted from each region of interest (ROI) in and around the ablation zone, including morphological, first-order, and transform-based features. These features were screened using LASSO regression and tenfold cross-validation algorithms. Several features were identified from the AZ, PAZ1, PAZ2, and PAZ3 radiomics models (Fig. 3), and feature values with weight coefficients greater than 0.4 in these models were selected to construct the AZ-PAZ model individually. These selected features were combined with clinical risk factors, including tumor size, to construct three clinical and radiomics-based combined models: the AZ-PAZ1 combined model (C1), the AZ-PAZ2 combined model (C2), and the AZ-PAZ3 combined model (C3).

Model Performances

Figure 4 shows the performance of each prediction model. In the training cohort, the PAZ2 model had a higher area under the curve (AUC) of 0.874 (95% CI, 0.776 to 0.973) in the single model (Table 2). In the combined AZ and PAZ model, the AUC was higher in the AZ-PAZ2 model at 0.896 (95% CI, 0.809 to 0.983) (Table 3). Among the combined clinical and radiomics models, the C2 model had a higher AUC of 0.919 (95% CI, 0.843 to 0.995) (Table 4). The DeLong's test showed that the difference between the C2 model and the clinical model in the training group was statistically significant (p = 0.041). However, there was no significant difference between the three C models (p > 0.05). As shown in Fig. 5, in the present study, the decision curve analysis results showed that the clinical utility of AZ-PAZ2 was superior to that of the Clinic, AZ, and PAZ2 models within a reasonable range of threshold probabilities. The C2 model obtained the ideal risk score thresholds via ROC curves of 0.741 and 0.699 for the training and validation sets, respectively. The thresholds classified all enrolled patients into two categories: high-risk groups greater than or equal to the thresholds, and low-risk groups less than the thresholds. The Kaplan-Meier curves showed that, in both the training and validation cohorts, the progression-free survival of the C2 model between the high-risk group and the low-risk group ( PFS) was significantly different (Fig. 6).

Table 2

Predictive performance of ablation zone and peri-ablation zone radiomics models for training and validation groups.
	Model	AZ	PAZ1	PAZ2	PAZ3
Training cohort	AUC	0.811(0.699,0.923)	0.845(0.725,0.964)	0.874(0.776,0.973)	0.848(0.731,0.964)
	Sensitivity	0.895	0.684	0.895	0.842
	Specificity	0.667	0.923	0.769	0.846
Validation cohort	AUC	0.794(0.612,0.976)	0.799(0.586,0.973)	0.831(0.638,1)	0.824(0.657,0.990)
	Sensitivity	0.750	0.875	0.875	0.875
	Specificity	0.824	0.647	0.824	0.706

Table 3

Predictive performance of ablation zone and peri-ablation zone combined radiomics models for training and validation groups.
	Model	AZ-PAZ1	AZ-PAZ2	AZ-PAZ3
Training cohort	AUC	0.872(0.783, 0.960)	0.896(0.809,0.983)	0.879(0.793,0.964)
	Sensitivity	0.947	0.895	0.895
	Specificity	0.744	0.846	0.795
Validation cohort	AUC	0.860(0.707,1)	0.882(0.747,1)	0.838(0.660,1)
	Sensitivity	0.750	1	0.875
	Specificity	0.824	0.647	0.765

Table 4

Predictive performance of clinical, clinical and radiomics combined radiomics models for training and validation groups.
	Model	Clinic	C1	C2	C3
Training cohort	AUC	0.763(0.626,0.900)	0.897(0.821,0.974)	0.919(0.843,0.995)	0.911(0.836,0.986)
	Sensitivity	0.769	0.667	0.846	0.769
	Specificity	0.579	1	0.895	0.842
Validation cohort	AUC	0.757(0.533,0.981)	0.853(0.699,1)	0.875(0.737,1)	0.831(0.671,0.990)
	Sensitivity	0.647	0.588	0.824	0.647
	Specificity	0.875	1	0.750	0.875
C1: clinical and AZ-PAZ1 combined model;
C2: clinical and AZ-PAZ2 combined model;
C3: clinical and AZ-PAZ3 combined model.

For patients diagnosed with CRLM, ablation and surgery are preferred treatment options. However, with the continuous improvement of comprehensive treatment, the treatment of CRLM is gradually moving towards interventional minimally invasive procedures, especially for patients who are not suitable for surgery. Ablation therapy is a minimally invasive treatment method that has gradually become popular in the treatment of CRLM. It can achieve effective treatment of small tumors without the need for major surgery and has the advantages of being less painful and reducing hospital stays^8–10,27. However, due to the specific characteristics of ablation therapy, there are certain risks involved in treatment, such as incomplete ablation or tumor recurrence²⁸. Therefore, accurate individualized risk assessment of local tumor progression (LTP) is crucial for monitoring protocols and therapeutic decision-making in CRLM patients undergoing ablation therapy. In this study, we developed a radiomics model for early prediction of LTP after ablation and demonstrated its high performance.

Due to the rich vascularity of the liver, the infiltration of intrahepatic malignant tumors into surrounding tissues via microvascular invasion is an important factor in tumor recurrence²⁹. Shan et al ³⁰ studied peri-tumoral 2cm radiomics features of hepatocellular carcinoma (HCC) based on CT images and found that they were more effective at predicting recurrence of HCC compared to radiomics features of the tumor. Xia et al ³¹ also found that tumors extracted from preoperative CT images through radiomic features of the surrounding region contributed to the microvascular invasion status of HCC. The peri-tumor area can be seen to provide effective information in predicting early recurrence. LTP was defined as the appearance of a new tumor within 10 mm of the tumor periphery on the image. Staal et al ³² evaluated that the radiomics features of the 10 mm margin around the postoperative AZ were also valuable markers for predicting LTP after thermal ablation of CRLM. To obtain complete and valid information around the AZ, we studied 0–5 mm, 0–10 mm, and 5–15 mm around the AZ. The best AUC was obtained for the radiomics model of 0–10 mm around the AZ, which is consistent with the definition of LTP. However, when assessing the dilatation around the AZ, it is not necessarily better to have a closer or farther outcome variable. We speculate that the 0–10 mm region around the AZ, representing the area where tumor molecules diffuse after ablation, may be the area with the greatest variation in capillary and immune factor content and the most diverse imaging information. Combining both AZ and PAZ showed superior efficacy in predicting early recurrence after CRLM ablation, with good sensitivity and specificity, helping clinicians to be more objective and individualized in evaluating early efficacy and designing treatment strategies for CRLM ablation.

The square root of GLCM_Correlation in both PAZ1 and PAZ2 models was one of the important features in their respective models with a weighting factor > 0.4. Previous studies have also found that the derived feature of GLCM_Correlation is an important predictor of recurrence after hepatectomy for CRLM^33,34, but based on CRLM lesions rather than PAZ. Our study, however, is based on PAZ. We speculate that this may represent the high recurrence rate due to the presence of tumor infiltration around the ablation zone and high malignancy. It is evident that some derived features based on GLCM_Correlation are important in predicting recurrence of some malignant tumors.

It has been shown that the type of thermal ablation does not determine the success and survival of ablation³⁵, so this study does not differentiate between ablation modalities. Lesions with a diameter greater than 3 cm have been shown to have significantly worse ablation results compared to lesions smaller than 3 cm³⁶. Therefore, only lesions with a diameter of less than 3 cm were included in this study. However, it was found that tumor size remains an independent predictor of local tumor progression (LTP) after thermal ablation in patients with CRLM, and the results showed that the larger the tumor, the greater the risk of LTP. This may be due to the fact that as the diameter of the lesion increases, the probability of tumor infiltration into the surrounding microvasculature increases, and the probability of complete ablation of the lesion decreases, leading to an increased risk of residual or recurrent tumor.

This study still has some limitations. Firstly, it is a single institution retrospective study with a small sample size, which may lead to some statistical bias and the inability to perform external validation analysis. It is necessary to further expand the sample size and validate the results. Secondly, the follow-up period of at least 6 months is rather short. Although local tumor progression usually occurs early in the follow-up period, and the majority of patients were followed up for > 12 months (37/53, 69%), some patients with short follow-up may still develop local tumor progression later. Finally, it was not possible to clearly interpret the impact of adjuvant and neoadjuvant chemotherapy before and after ablation on preventing local tumor progression in patients with CRLM.

In conclusion, the AZ, PAZ, and combined models based on post-ablation MRI images of CRLM patients can early predict LTP after ablation treatment, especially within a radius of 10 mm around the ablation zone, which provides valuable guidance for individualized treatment, follow-up, and evaluation strategies for CRLM patients.

Data availability

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

Authors contributions

Xiucong Zhu and Zhenhua Zhao conceived of the study. Jinke Zhu participated in the design of the study and the initial syntheses of results. Xiucong Zhu, Chenwen Sun and Fandong Zhu prepared figures and wrote the manuscript text. Xiucong Zhu collected the follow-up data. Fandong Zhu and Chenwen Sun made statistical analysis. All authors read and approved the final manuscript.

Competing interests

The authors declare no competing interests.

Sung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 71, 209-249, doi:10.3322/caac.21660 (2021).
Ganesh, K. et al. Immunotherapy in colorectal cancer: rationale, challenges and potential. Nature Reviews Gastroenterology & Hepatology 16, 361-375, doi:10.1038/s41575-019-0126-x (2019).
Engstrand, J., Nilsson, H., Stromberg, C., Jonas, E. & Freedman, J. Colorectal cancer liver metastases - a population-based study on incidence, management and survival. BMC Cancer 18, 78, doi:10.1186/s12885-017-3925-x (2018).
van der Geest, L. G. et al. Nationwide trends in incidence, treatment and survival of colorectal cancer patients with synchronous metastases. Clin Exp Metastasis 32, 457-465, doi:10.1007/s10585-015-9719-0 (2015).
van de Velde, C. J. et al. EURECCA colorectal: multidisciplinary management: European consensus conference colon & rectum. Eur J Cancer 50, 1 e1-1 e34, doi:10.1016/j.ejca.2013.06.048 (2014).
Knott, E. A. et al. Microwave ablation for colorectal cancer metastasis to the liver: a single-center retrospective analysis. Journal of Gastrointestinal Oncology 12, 1454-1469, doi:10.21037/jgo-21-159 (2021).
Cummings, L. C., Payes, J. D. & Cooper, G. S. Survival after hepatic resection in metastatic colorectal cancer: a population-based study. Cancer 109, 718-726, doi:10.1002/cncr.22448 (2007).
Takahashi, H. & Berber, E. Role of thermal ablation in the management of colorectal liver metastasis. Hepatobiliary Surg Nutr 9, 49-58, doi:10.21037/hbsn.2019.06.08 (2020).
Puijk, R. S. et al. Improved Outcomes of Thermal Ablation for Colorectal Liver Metastases: A 10-Year Analysis from the Prospective Amsterdam CORE Registry (AmCORE). Cardiovasc Intervent Radiol 45, 1074-1089, doi:10.1007/s00270-022-03152-9 (2022).
Qin, S. et al. The local efficacy and influencing factors of ultrasound-guided percutaneous microwave ablation in colorectal liver metastases: a review of a 4-year experience at a single center. International Journal of Hyperthermia 36, 36-43, doi:10.1080/02656736.2018.1528511 (2018).
Wang, C.-Z., Yan, G.-X., Xin, H. & Liu, Z.-Y. Oncological outcomes and predictors of radiofrequency ablation of colorectal cancer liver metastases. World Journal of Gastrointestinal Oncology 12, 1044-1055, doi:10.4251/wjgo.v12.i9.1044 (2020).
Puijk, R. S. et al. Colorectal liver metastases: surgery versus thermal ablation (COLLISION) - a phase III single-blind prospective randomized controlled trial. BMC Cancer 18, 821, doi:10.1186/s12885-018-4716-8 (2018).
Liu, M. et al. Percutaneous thermal ablation for the treatment of colorectal liver metastases and hepatocellular carcinoma: a comparison of local therapeutic efficacy. International Journal of Hyperthermia 33, 446-453, doi:10.1080/02656736.2017.1278622 (2017).
Kurilova, I. et al. Factors Associated With Local Tumor Control and Complications After Thermal Ablation of Colorectal Cancer Liver Metastases: A 15-year Retrospective Cohort Study. Clinical Colorectal Cancer 20, e82-e95, doi:10.1016/j.clcc.2020.09.005 (2021).
Shady, W. et al. Percutaneous Microwave versus Radiofrequency Ablation of Colorectal Liver Metastases: Ablation with Clear Margins (A0) Provides the Best Local Tumor Control. Journal of Vascular and Interventional Radiology 29, 268-275.e261, doi:10.1016/j.jvir.2017.08.021 (2018).
Schulz, A. et al. Diagnostic performance of CT, MRI and PET/CT in patients with suspected colorectal liver metastases: the superiority of MRI. Acta Radiologica 57, 1040-1048, doi:10.1177/0284185115617349 (2016).
Sivesgaard, K. et al. Diagnostic accuracy of CE-CT, MRI and FDG PET/CT for detecting colorectal cancer liver metastases in patients considered eligible for hepatic resection and/or local ablation. European radiology 28, 4735-4747, doi:10.1007/s00330-018-5469-0 (2018).
Schnitzer, M. L. et al. Economic evaluation of 18F-FDG PET/CT, MRI and CE-CT in selection of colorectal liver metastases eligible for ablation – A cost-effectiveness analysis. European journal of radiology 163, doi:10.1016/j.ejrad.2023.110803 (2023).
Lambin, P. et al. Radiomics: the bridge between medical imaging and personalized medicine. Nature Reviews Clinical Oncology 14, 749-762, doi:10.1038/nrclinonc.2017.141 (2017).
Wu, Y., Ma, Q., Fan, L., Wu, S. & Wang, J. An Automated Breast Volume Scanner-Based Intra- and Peritumoral Radiomics Nomogram for the Preoperative Prediction of Expression of Ki-67 in Breast Malignancy. Academic radiology 31, 93-103, doi:10.1016/j.acra.2023.07.004 (2024).
Zhang, X. et al. Radiomics nomogram based on multi-parametric magnetic resonance imaging for predicting early recurrence in small hepatocellular carcinoma after radiofrequency ablation. Frontiers in oncology 12, doi:10.3389/fonc.2022.1013770 (2022).
Zhu, F. et al. CT-based radiomics models may predict the early efficacy of microwave ablation in malignant lung tumors. Cancer Imaging 23, doi:10.1186/s40644-023-00571-w (2023).
Koska, I. O. et al. Radiomics in differential diagnosis of Wilms tumor and neuroblastoma with adrenal location in children. European radiology 34, 5016-5027, doi:10.1007/s00330-024-10589-8 (2024).
Shahveranova, A. et al. Prediction of Local Tumor Progression After Microwave Ablation in Colorectal Carcinoma Liver Metastases Patients by MRI Radiomics and Clinical Characteristics-Based Combined Model: Preliminary Results. Cardiovasc Intervent Radiol 46, 713-725, doi:10.1007/s00270-023-03454-6 (2023).
Shang, Y. et al. Computed Tomography-derived intratumoral and peritumoral radiomics in predicting EGFR mutation in lung adenocarcinoma. La radiologia medica 128, 1483-1496, doi:10.1007/s11547-023-01722-6 (2023).
Jiang, W. et al. Intra‐ and Peritumoral Based Radiomics for Assessment of Lymphovascular Invasion in Invasive Breast Cancer. Journal of Magnetic Resonance Imaging 59, 613-625, doi:10.1002/jmri.28776 (2023).
Crocetti, L., Scalise, P., Bozzi, E., Candita, G. & Cioni, R. Thermal ablation of hepatocellular carcinoma. Journal of Medical Imaging and Radiation Oncology 67, 817-831, doi:10.1111/1754-9485.13613 (2023).
Yang, G. et al. The prognosis of radiofrequency ablation versus hepatic resection for patients with colorectal liver metastases: A systematic review and meta-analysis based on 22 studies. Int J Surg 87, 105896, doi:10.1016/j.ijsu.2021.105896 (2021).
Li, J. et al. Preoperative prediction and risk assessment of microvascular invasion in hepatocellular carcinoma. Critical Reviews in Oncology/Hematology 190, doi:10.1016/j.critrevonc.2023.104107 (2023).
Shan, Q.-y. et al. CT-based peritumoral radiomics signatures to predict early recurrence in hepatocellular carcinoma after curative tumor resection or ablation. Cancer Imaging 19, doi:10.1186/s40644-019-0197-5 (2019).
Xia, T.-y. et al. Predicting Microvascular Invasion in Hepatocellular Carcinoma Using CT-based Radiomics Model. Radiology 307, doi:10.1148/radiol.222729 (2023).
Staal, F. C. R. et al. Predicting local tumour progression after ablation for colorectal liver metastases: CT-based radiomics of the ablation zone. European journal of radiology 141, 109773, doi:10.1016/j.ejrad.2021.109773 (2021).
Granata, V. et al. EOB-MR Based Radiomics Analysis to Assess Clinical Outcomes following Liver Resection in Colorectal Liver Metastases. Cancers 14, doi:10.3390/cancers14051239 (2022).
Granata, V. et al. Radiomics textural features by MR imaging to assess clinical outcomes following liver resection in colorectal liver metastases. La radiologia medica 127, 461-470, doi:10.1007/s11547-022-01477-6 (2022).
Krul, M. F. et al. Radiofrequency versus microwave ablation for intraoperative treatment of colorectal liver metastases. European Journal of Surgical Oncology 48, 834-840, doi:10.1016/j.ejso.2021.10.012 (2022).
Tomita, K. et al. Evidence on percutaneous radiofrequency and microwave ablation for liver metastases over the last decade. Japanese Journal of Radiology 40, 1035-1045, doi:10.1007/s11604-022-01335-5 (2022).

No competing interests reported.

Download PDF

Reviews received at journal
08 Oct, 2024
Reviewers agreed at journal
02 Oct, 2024
Reviewers invited by journal
01 Oct, 2024
Editor assigned by journal
19 Sep, 2024
Editor invited by journal
21 Aug, 2024
Submission checks completed at journal
21 Aug, 2024
First submitted to journal
11 Aug, 2024

You are reading this latest preprint version

Early prediction of local tumor progression after ablation of colorectal liver metastases based on MRI radiomics

Status:

Version 1

Abstract

Figures

Introduction

Materials and methods

Patients

Ablation Procedure

Radiomics Analysis

Image Acquisition

Image Segmentation

Feature Extraction

Radiomic feature selection and model building

Statistical analysis

Result

Clinical Characteristics

Radiomics Features

Model Performances

Discussion

Conclusion

Declarations

References

Additional Declarations

Status:

Version 1