Long ZD, Yu X, Xing ZX, Wang R. Multiparameter magnetic resonance imaging-based radiomics model for the prediction of rectal cancer metachronous liver metastasis. World J Gastrointest Oncol 2025; 17(1): 96598 [DOI: 10.4251/wjgo.v17.i1.96598]
Corresponding Author of This Article
Rui Wang, MD, Doctor, Department of Hepatobiliary and Pancreaticosplenic Surgery, Jingzhou Hospital Affiliated to Yangtze University, No. 26 Chuyuan Road, Jingzhou District, Jingzhou 434100, Hubei Province, China. hyhq0216@163.com
Research Domain of This Article
Gastroenterology & Hepatology
Article-Type of This Article
Retrospective Cohort Study
Open-Access Policy of This Article
This article is an open-access article which was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Zhi-Da Long, Xiao Yu, Zhi-Xiang Xing, Rui Wang, Department of Hepatobiliary and Pancreaticosplenic Surgery, Jingzhou Hospital Affiliated to Yangtze University, Jingzhou 434100, Hubei Province, China
Author contributions: Wang R designed the research study; Long ZD has completed the preliminary data collection and visualization analysis; Yu X and Xing ZX have completed the initial draft and proofreading of their paper; All authors have made final corrections to the manuscript.
Institutional review board statement: This study has been approved by the Ethics Committee of Jingzhou Central Hospital and complies with the Helsinki Declaration. All included patients were exempt from informed consent, No. 2024-154-01.
Informed consent statement: As the study only involved retrospective chart reviews, informed written consents were not required in accordance with institutional IRB policy.
Conflict-of-interest statement: The authors declare that they have no conflict of interest.
Data sharing statement: Not applicable.
STROBE statement: The authors have read the STROBE Statement—a checklist of items, and the manuscript was prepared and revised according to the STROBE Statement-a checklist of items.
Open-Access: This article is an open-access article that was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution NonCommercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: https://creativecommons.org/Licenses/by-nc/4.0/
Corresponding author: Rui Wang, MD, Doctor, Department of Hepatobiliary and Pancreaticosplenic Surgery, Jingzhou Hospital Affiliated to Yangtze University, No. 26 Chuyuan Road, Jingzhou District, Jingzhou 434100, Hubei Province, China. hyhq0216@163.com
Received: May 10, 2024 Revised: September 6, 2024 Accepted: September 27, 2024 Published online: January 15, 2025 Processing time: 215 Days and 20.5 Hours
Abstract
BACKGROUND
The liver, as the main target organ for hematogenous metastasis of colorectal cancer, early and accurate prediction of liver metastasis is crucial for the diagnosis and treatment of patients. Herein, this study aims to investigate the application value of a combined machine learning (ML) based model based on the multiparameter magnetic resonance imaging for prediction of rectal metachronous liver metastasis (MLM).
AIM
To investigate the efficacy of radiomics based on multiparametric magnetic resonance imaging images of preoperative first diagnosed rectal cancer in predicting MLM from rectal cancer.
METHODS
We retrospectively analyzed 301 patients with rectal cancer confirmed by surgical pathology at Jingzhou Central Hospital from January 2017 to December 2023. All participants were randomly assigned to the training or validation queue in a 7:3 ratio. We first apply generalized linear regression model (GLRM) and random forest model (RFM) algorithm to construct an MLM prediction model in the training queue, and evaluate the discriminative power of the MLM prediction model using area under curve (AUC) and decision curve analysis (DCA). Then, the robustness and generalizability of the MLM prediction model were evaluated based on the internal validation set between the validation queue groups.
RESULTS
Among the 301 patients included in the study, 16.28% were ultimately diagnosed with MLM through pathological examination. Multivariate analysis showed that carcinoembryonic antigen, and magnetic resonance imaging radiomics were independent predictors of MLM. Then, the GLRM prediction model was developed with a comprehensive nomogram to achieve satisfactory differentiation. The prediction performance of GLRM in the training and validation queue was 0.765 [95% confidence interval (CI): 0.710-0.820] and 0.767 (95%CI: 0.712-0.822), respectively. Compared with GLRM, RFM achieved superior performance with AUC of 0.919 (95%CI: 0.868-0.970) and 0.901 (95%CI: 0.850-0.952) in the training and validation queue, respectively. The DCA indicated that the predictive ability and net profit of clinical RFM were improved.
CONCLUSION
By combining multiparameter magnetic resonance imaging with the effectiveness and robustness of ML-based predictive models, the proposed clinical RFM can serve as an insight tool for preoperative assessment of MLM risk stratification and provide important information for individual diagnosis and treatment of rectal cancer patients.
Core Tip: In recent years, with the rapid development of data and information technology, imaging omics has been gradually applied in the clinical diagnosis and treatment of tumors, as it can non-invasive extract high-throughput heterogeneity information within tumors and integrate patient clinical information to improve the accuracy of models. Up to now, imaging omics models based on computed tomography or magnetic resonance imaging (MRI) images have shown potential application value in preoperative T and N staging and efficacy evaluation of rectal cancer. However, there is currently very little imaging omics research based on MRI of primary rectal cancer tumors. In fact, MRI is the most accurate imaging method for diagnosing rectal cancer, which can better display the invasion of adjacent lymph nodes, blood vessels, or surrounding organs by primary rectal cancer tumors. In view of this, this study attempts to establish a non-invasive preoperative prediction model for metachronous liver metastasis in rectal cancer based on the imaging omics features of the initial diagnosis MRI images of rectal cancer, combined with machine learning algorithms, and verify its effectiveness. This will provide clinical assistance for clinicians to make personalized monitoring and treatment decision.
Citation: Long ZD, Yu X, Xing ZX, Wang R. Multiparameter magnetic resonance imaging-based radiomics model for the prediction of rectal cancer metachronous liver metastasis. World J Gastrointest Oncol 2025; 17(1): 96598
Throughout the world, colorectal cancer is a common malignant tumor of the digestive system, with its incidence rate and mortality ranking top[1]. According to the statistics of global cancer statistics, the global incidence rate and mortality of colorectal cancer in 2020 will be 10% and 9.4% respectively[2-4]. The liver, as the most common target organ for hematogenous metastasis of gastrointestinal malignant tumors, is also the most susceptible site for rectal cancer metastasis[5]. Previous studies have shown that approximately 35% to 55% of rectal cancer patients experience liver metastasis, and ultimately about 40% to 50% of rectal cancer patients die from liver metastasis[6-8]. At present, surgical resection of liver metastases is the only possible method for rectal cancer patients with liver metastases to achieve cure or prolong survival, and patients who can accept radical resection of liver metastases have a significantly higher 5-year survival rate than those who cannot be completely removed[9,10]. In addition, the more 15% to 25% of rectal cancer patients are diagnosed or undergo rectal cancer radical surgery without liver metastasis, the more likely they may eventually develop rectal metachronous liver metastasis (MLM) as the condition progresses[11,12]. Therefore, early detection and prediction of liver metastasis in rectal cancer and effective clinical intervention are crucial for the survival and prognosis of rectal cancer patients.
Up to now, there is no effective research method that can accurately predict the MLM of rectal cancer. There are studies based on computed tomography (CT) images of liver parenchyma to predict MLM by obtaining intensity features (such as grayscale values, entropy values, etc.), but the clinical promotion value is limited[13-15]. Previous researchers have attempted to analyze the association between clinical baseline features (such as tumor markers, lymph node metastasis status, etc.) and liver metastasis in rectal cancer, but there is currently no consensus or clear understanding of the relationship between MLM and clinical pathological parameters[16,17]. In addition, some scholars have studied the impact of gene mutations such as KRAS/NRAS on MLM, but genomics is an invasive examination that is expensive and difficult to widely promote in rectal cancer patients[18]. In view of this, it is urgent to seek convenient and high-precision prediction models that can be used to predict MLM to assist clinical decision-making.
In recent years, with the rapid development of data and information technology, imaging omics has been gradually applied in the clinical diagnosis and treatment of tumors, as it can non-invasive extract high-throughput heterogeneity information within tumors and integrate patient clinical information to improve the accuracy of models. Up to now, imaging omics models based on CT or magnetic resonance imaging (MRI) images have shown potential application value in preoperative T and N staging and efficacy evaluation of rectal cancer[19,20]. However, there is currently very little imaging omics research based on MRI of primary rectal cancer tumors. In fact, MRI is the most accurate imaging method for diagnosing rectal cancer, which can better display the invasion of adjacent lymph nodes, blood vessels, or surrounding organs by primary rectal cancer tumors. In view of this, this study attempts to establish a non-invasive preoperative prediction model for MLM in rectal cancer based on the imaging omics features of the initial diagnosis MRI images of rectal cancer, combined with machine learning (ML) algorithms, and verify its effectiveness. This will provide clinical assistance for clinicians to make personalized monitoring and treatment decision.
MATERIALS AND METHODS
Patients
We retrospectively analyzed 301 patients with rectal cancer confirmed by surgical pathology at Jingzhou Central Hospital from January 2017 to December 2023. The inclusion criteria are as follows: (1) Patients confirmed by postoperative histopathology to have rectal adenocarcinoma; (2) Patients with regular follow-up before and after treatment, and complete follow-up imaging and clinical data; (3) Patients who were diagnosed with rectal cancer for the first time but did not find distant metastasis on enhanced CT or MRI; and (4) Patients who have no history of other primary malignant tumors and have not received anti-tumor treatment before their initial diagnosis. Exclusion criteria: (1) Patients confirmed by histopathology to have rectal mucinous adenocarcinoma (i.e., due to the high malignancy of rectal mucinous adenocarcinoma and potential risk of metastasis); and (2) Patients with poor MR image quality or severe artifacts that cannot meet the analysis requirements. This study has been approved by the Ethics Committee of Jingzhou Central Hospital and complies with the Helsinki Declaration. All included patients were exempt from informed consent (No. 2024-154-01). The process and predictive model construction for patients included in the study were summarized in Figure 1.
Figure 1 Patient inclusion and prediction model construction process.
MLM: Metachronous liver metastasis.
Acquisition of rectal MRI images
Region of interest (ROI) drawing and radiomics feature extraction: A radiologist with more than 5 years of experience in abdominal and pelvic MRI diagnosis refers to all sequence images to determine the specific location and boundary of the primary tumor of rectal cancer on high-definition T2 weighted imaging (HD T2WI), diffusion weighted imaging (DWI) (b = 1000) The region of interest was drawn layer by layer on the image and reviewed by a senior radiologist with more than 10 years of experience in abdominal and pelvic MRI diagnosis. The ROIs of all layers of the entire tumor are fused to form a three-dimensional morphology of the tumor. Finally, the minimum absolute shrinkage and selection operator algorithm is used to extract and screen radiomic features related to MLM (including first-order features, morphological features, and second-order texture features, high-order texture features).
MRI imaging data collection and analysis
As shown in Supplementary Figure 1, the DWI (b = 1000) and oblique axial HD T2WI images in MRI of patients with rectal cancer were imported into ITK-SNAP software. We used the Z-Score standard to resample and normalize the original data to achieve image resampling and normalization. The gray value of the image is adjusted to the standard normal distribution. At the same time, consistency analysis is performed before screening radiomics features to eliminate radiomics features with poor consistency.
Construction and validation of imaging-based radiomics MLM model
We adopted two commonly used ML algorithm prediction models in this MLM prediction, namely: Generalized linear regression model (GLRM) and random forest model (RFM). In addition, we also used the least absolute shrinkage and selection operator (LASSO) algorithm to reduce the dimensionality of the radiomics features extracted from the training set, and then used logistic regression analysis on the reduced feature set to construct radiomics labels and obtain each patient label score to reflect the risk of MLM. Screen radiomic features related to MLM. The prediction performance of the prediction model in the training set and validation set was evaluated using the receiver operating characteristic curve [i.e., accuracy, sensitivity, specificity, and area under the curve (AUC)] and decision curve analysis. The DeLong test was used to compare AUCs between prediction models to select the radiomics prediction model with the best prediction performance.
Statistical analysis
In this study, statistical analysis and visualization of clinical baseline characteristics of patients between the MLM group and the non-MLM group were implemented using R software (version 4.2.3). All comparisons involving categorical variables (i.e., percentile counts) were performed using the χ2 test. If the continuous variables (i.e., mean ± SD) conformed to the normal distribution, the t test was used. If they did not conform to the normal distribution, the Mann-Whitney U test was used. P less than 0.05 was considered as statistically significant difference between groups.
RESULTS
Patients’ characteristics and radiomics features
Our study included 301 rectal cancer patients who underwent MR examination before treatment. Among them, 49 patients developed liver metastasis within 24 months after diagnosis of rectal cancer, including 22 males and 27 females. The average time for liver metastasis was 8.61 ± 6.50 months. In addition, the median follow-up time for all patients was 23.5 months. We also extracted 40 radiomics features from each patient's HD T2WI and DWI images, and ultimately found that there were five radiomics features compared between the MLM and nMLM groups [i.e. T2_wavelet.LHL_ firstorder_RobustMeanAbsoluteDeviation (Character 1), T1_log.sigma.1.0.mm.3D_firstorder_Kurtosis (Character 2), ADC_wavelet.LHL_gldm_LargeDepenceHighGrayLevelEmphasis (Character 4), ADC_wavelet.HLL_glrlm_LongRunHighGrayLeveEmphasis (Character 5), T1_wavelet.HHL_glszm_GrayLevelNonUniformity (Character 7)]. The clinical baseline characteristics of all patients are summarized in Table 1 and Supplementary Table 1.
Table 1 Patient baseline characteristics and magnetic resonance imaging radiomics parameters, n (%).
Variables
Overall (n = 301)
MLM (n = 49)
Non-MLM (n = 252)
P value
Age [median (IQR)]
46.00 (31.00, 58.00)
51.00 (32.00, 60.00)
44.50 (30.00, 57.00)
0.074
Gender
Male
136 (45.2)
22 (44.9)
114 (45.2)
1.000
Female
165 (54.8)
27 (55.1)
138 (54.8)
BMI, kg/m2 [median (IQR)]
23.50 (20.80, 26.10)
22.90 (20.70, 26.60)
23.55 (20.80, 26.02)
0.870
Smoking
Yes
140 (46.5)
26 (53.1)
114 (45.2)
0.396
No
161 (53.5)
23 (46.9)
138 (54.8)
Drinking
Yes
142 (47.2)
22 (44.9)
120 (47.6)
0.847
No
159 (52.8)
27 (55.1)
132 (52.4)
MTD, cm
≤ 4
147 (48.8)
25 (51.0)
122 (48.4)
0.859
> 4
154 (51.2)
24 (49.0)
130 (51.6)
Pathological grade
I-II
142 (47.2)
17 (34.7)
125 (49.6)
0.079
III
159 (52.8)
32 (65.3)
127 (50.4)
Pathological type
Infiltrating
38 (12.6)
4 (8.2)
34 (13.5)
0.700
Ulcerative
111 (36.9)
18 (36.7)
93 (36.9)
Cauliflower
74 (24.6)
12 (24.5)
62 (24.6)
Uplift
78 (25.9)
15 (30.6)
63 (25.0)
MRF
Yes
162 (53.8)
23 (46.9)
139 (55.2)
0.368
No
139 (46.2)
26 (53.1)
113 (44.8)
EMVI
Yes
154 (51.2)
26 (53.1)
128 (50.8)
0.893
No
147 (48.8)
23 (46.9)
124 (49.2)
Node
Yes
133 (44.2)
26 (53.1)
107 (42.5)
0.226
No
168 (55.8)
23 (46.9)
145 (57.5)
Spicule
Yes
164 (54.5)
28 (57.1)
136 (54.0)
0.801
No
137 (45.5)
21 (42.9)
116 (46.0)
AFP, ng/mL [median (IQR)]
2.81 (2.25, 3.41)
2.69 (2.16, 3.32)
2.82 (2.27, 3.42)
0.334
CA199, U/mL [median (IQR)]
9.61 (6.60, 12.66)
9.83 (6.64, 11.47)
9.59 (6.54, 12.73)
0.856
CEA, ng/mL [median (IQR)]
0-5
167 (55.5)
14 (28.6)
153 (60.7)
< 0.001
≥ 5
134 (44.5)
35 (71.4)
99 (39.3)
pT stage
T2
158 (52.5)
30 (61.2)
128 (50.8)
0.237
T3
143 (47.5)
19 (38.8)
124 (49.2)
N stage
N0
94 (31.2)
16 (32.7)
78 (31.0)
0.936
N1
102 (33.9)
17 (34.7)
85 (33.7)
N2
105 (34.9)
16 (32.7)
89 (35.3)
Character 1 [median (IQR)]
0.46 (0.31, 0.63)
1.65 (1.44, 1.98)
0.41 (0.28, 0.55)
< 0.001
Character 2 [median (IQR)]
8.70 (7.70, 9.60)
13.00 (11.80, 14.80)
8.30 (7.57, 9.10)
< 0.001
Character 3 [median (IQR)]
169.30 (140.90, 198.10)
167.80 (145.20, 194.70)
172.25 (139.95, 198.25)
0.976
Character 4 [median (IQR)]
1.70 (1.20, 2.40)
4.50 (3.60, 5.10)
1.60 (1.10, 2.10)
< 0.001
Character 5 [median (IQR)]
3.60 (2.90, 4.20)
6.70 (4.80, 8.80)
3.40 (2.70, 4.00)
< 0.001
Character 6 [median (IQR)]
32.30 (22.30, 42.50)
29.80 (23.00, 39.30)
32.75 (22.08, 43.00)
0.364
Character 7 [median (IQR)]
4.20 (2.80, 5.30)
7.20 (5.90, 8.10)
3.75 (2.60, 4.80)
< 0.001
Candidate predictive factors selection related to MLM
We applied LASSO analysis to obtain the optimal feature subset from candidate MRI imaging parameters, and finally screen five imaging parameter candidate variables based on the parameter weights in the subset, namely feature 1, feature 2, feature 4, feature 5, and feature 7 (Figure 2). Additionally, multivariate logistic regression analysis was conducted to determine independent risk factors for MLM. The results showed that feature 1 [odds ratio (OR) = 1.19; 95% confidence interval (CI): 0.87-2.16], feature 2 (OR = 2.58; 95%CI: 1.14-3.99), feature 4 (OR = 1.13; 95%CI: 0.61-1.94), feature 5 (OR = 0.91; 95%CI: 0.19-2.06), and feature 7 (OR = 2.09; 95%CI: 1.05-3.55) were significantly associated with the occurrence of MLM (Table 2).
Figure 2 Selection of metachronous liver metastasis candidate predictive variables based on least absolute shrinkage and selection operator regression.
A: Spearman correlation analysis; B: Least absolute shrinkage and selection operator regression analysis. BMI: Body mass index; MTD: Maximum tumor diameter; MRF: Mesenteric fascia; AFP: Alpha fetoprotein; CA: Carbohydrate antigen; CEA: Carcinoembryonic antigen.
Table 2 Selection of metachronous liver metastasis predictive variables based on logistic regression.
Variables
Univariate
Multivariate
OR
95%CI
P value
OR
95%CI
P value
Age [median (IQR)]
1.02
0.55-1.72
> 0.05
Gender
Male
1.00
Female
0.86
0.23-1.02
> 0.05
BMI, kg/m2 [median (IQR)]
1.21
0.78-1.87
> 0.05
Smoking
Yes
1.00
No
0.79
0.18-1.16
> 0.05
Drinking
Yes
1.00
No
0.81
0.23-2.57
> 0.05
MTD, cm
≤ 4
1.00
> 4
1.16
0.77-3.98
> 0.05
Pathological grade
I-II
1.00
III
1.23
0.91-2.26
> 0.05
Pathological type
Infiltrating
1.00
Ulcerative
0.63
0.12-1.16
> 0.05
Cauliflower
0.78
0.23-1.65
> 0.05
Uplift
1.12
0.88-2.37
> 0.05
MRF
Yes
1.00
No
0.63
0.22-1.19
> 0.05
EMVI
Yes
1.00
No
0.67
0.11-1.59
> 0.05
Node
Yes
1.00
No
0.82
0.41-2.16
> 0.05
Spicule
Yes
1.00
No
0.58
0.21-2.26
> 0.05
AFP, ng/mL [median (IQR)]
1.23
0.88-1.77
> 0.05
CA199, U/mL [median (IQR)]
1.15
0.91-1.65
> 0.05
CEA, ng/mL [median (IQR)]
2.13
1.03-4.57
< 0.05
2.11
1.01-4.12
< 0.05
pT stage
T2
1.00
T3
1.13
0.75-2.16
> 0.05
N stage
N0
1.00
N1
1.13
0.86-1.87
> 0.05
N2
1.21
0.99-2.14
> 0.05
Character 1 [median (IQR)]
1.23
0.77-1.96
< 0.05
1.19
0.87-2.16
< 0.05
Character 2 [median (IQR)]
2.67
1.12-4.13
< 0.05
2.58
1.14-3.99
< 0.05
Character 3 [median (IQR)]
1.58
0.88-2.03
> 0.05
Character 4 [median (IQR)]
1.06
0.56-1.87
< 0.05
1.13
0.61-1.94
< 0.05
Character 5 [median (IQR)]
0.98
0.23-1.56
< 0.05
0.91
0.19-2.06
< 0.05
Character 6 [median (IQR)]
1.87
1.02-2.76
> 0.05
Character 7 [median (IQR)]
2.23
1.11-3.87
< 0.05
2.09
1.05-3.55
< 0.05
Construction and evaluation of nomogram predictive model for MLM
Based on the independent risk factors for MLM mentioned above, a nomogram prediction model was established (Figure 3). In the training queue, the AUC was 0.765 (95%CI: 0.710-0.820), with a sensitivity of 0.75 and a specificity of 0.98. In the validation queue, the AUC was 0.767 (95%CI: 0.712-0.822), with a sensitivity of 0.64 and a specificity of 0.94 (Table 3). The calibration curve showed good consistency between the predicted probability and the actual probability, indicating that the nomogram has good predictive performance.
Figure 3 Nomogram visual prediction model for predicting metachronous liver metastasis.
A: Nomogram; B: Calibration curve. CEA: Carcinoembryonic antigen.
Table 3 Evaluation of predictive performance of metachronous liver metastasis prediction model based on receiver operating characteristic.
Prediction model
Training set
International set
AUC
95%CI
PPV
NPV
AUC
95%CI
PPV
NPV
RFM
0.919
0.868-0.970
0.94
0.99
0.901
0.850-0.952
0.94
0.99
GLRM
0.765
0.710-0.820
0.75
0.98
0.767
0.712-0.822
0.64
0.94
Construction and evaluation of random forest predictive model based on ML algorithm
In the RFM, according to the importance ranking of predictive variable characteristics, feature 1, feature 2, feature 4, and feature 7 were identified as the four major important variables, while other clinical indices followed closely behind. Besides, the AUC of the training queue reached 0.919 (95%CI: 0.868-0.970), with a sensitivity of 0.94 and a specificity of 0.99 (Figure 4), significantly better than the AUC of traditional GLRM (P < 0.05). In the validation queue, the AUC of the RFM reached 0.901 (95%CI: 0.850-0.952), with a sensitivity of 0.94 and a specificity of 0.99. As shown in Figure 5, in terms of evaluating the predictive robustness of the model, RFM also demonstrates superior predictive performance compared to GLRM.
Figure 4 Random forest prediction model for predicting metachronous liver metastasis.
A: The random forest prediction model based on machine learning algorithms; B: Predictive performance detection of models. The red dots represent patients with metachronous liver metastasis, and the blue dots represent patients without metachronous liver metastasis.
Figure 5 Performance evaluation of predictive models based on decision curve analysis.
A: Training cohort; B: Testing cohort. RFM: Random forest model; GLRM: Generalized linear regression model.
Meanwhile, we used SHapley additive explanations to evaluate the RFM, as shown in Supplementary Figure 2. It was also observed that feature 1, feature 2, feature 4, and feature 7 play the most crucial roles in predicting and interpreting RFM. Specifically, elasticity score, maximum diameter and Adler blood flow grading were associated with an increased risk of MLM.
DISCUSSION
Surgery is currently one of the main treatments for colorectal liver metastases[21,22]. However, most patients with colorectal liver metastases have missed the optimal treatment time when diagnosed, resulting in poor survival and poor prognosis[9,23]. Our study attempts to construct and validate a fusion radiomics MLM prediction model from the MRI images of the first diagnosed primary tumors of patients with rectal cancer (i.e., high b-value DWI, HD T2WI), combined with the patient's clinical baseline characteristics and non-radiomics MRI features. The results suggest that compared with a single clinical predictor, the fused radiomics model has better predictive performance, suggesting that this prediction model has the potential to help clinicians adjust treatment plans based on the probability of liver metastasis, thereby improving the prognosis and long-term survival rate of rectal cancer patients.
Early detection of liver metastasis in rectal cancer has always been a major challenge in clinical diagnosis and treatment, especially in accurately predicting MLM, which is crucial for adjusting treatment plans reasonably[24-26]. However, there is currently no effective method in clinical practice. This study found that MRI has unparalleled advantages in multi parameter and multi plane imaging compared to other imaging examinations, and has become the preferred imaging method for tumor staging and re staging of rectal cancer patients after treatment. However, there are still few reports on predicting MLM based on MRI radiomics[27,28]. Previous studies reported using convolutional networks to analyze preoperative liver CT images of patients with rectal cancer, extract and analyze liver imaging features, and construct MLM predictions based on patient clinical characteristics[29-31]. Consistent with previous research, the GLRM model finally achieved AUCs of 0.765 and 0.767 in the validation set and training set respectively.
The results of this study also showed that the AUC of the RFM model built based on the radiomics and clinical baseline characteristics of HD T2WI and DWI of the primary tumor in the training set and validation set were 0.919 and 0.901 respectively, suggesting that MRI-based radiomics is effective for MLM. Unparalleled predictive advantage. We speculate that the high-definition T2WI sequence in the MRI sequence has the characteristics of small field of view, large matrix, high spatial resolution, and multiple signal acquisition times. Therefore, it can clearly display the invasion of adjacent structures by the tumor and help distinguish subtle differences, especially high b-value DWI sequences can show the degree of diffusion restriction of free water molecules in the tumor cell microenvironment and can reflect more heterogeneous information within the tumor.
Our study still inevitably has the following limitations. Firstly, this study is based on MRI images and clinical data of rectal cancer patients obtained from a single center institution. Subsequent studies still need to expand the sample size or further evaluate the MLM prediction performance of the fusion prediction model in conjunction with multi center institutions, as well as obtain external queue validation; Secondly, due to limitations in case imaging data, this study only obtained plain MRI images of the rectum at the initial diagnosis of rectal cancer patients, and there is still a lack of comparison between enhanced MRI images of the rectum. Therefore, subsequent studies need to add imaging omics features in dynamic enhanced sequences; Thirdly, this study is a retrospective study. Due to incomplete follow-up data of some rectal cancer patients, the results may be compromised. Therefore, future studies still need to focus on prospective cohort studies to reduce data bias errors.
CONCLUSION
In general, a radiomics prediction model based on preoperative primary tumor MRI images of rectal cancer patients combined with relevant clinical baseline features has high accuracy in predicting MLM of rectal cancer. In particular, RFM is expected to provide auxiliary guidance for clinical and imaging physicians to monitor MLM of rectal cancer, and help provide effective information for patients to develop personalized treatment plans, thereby helping to improve the long-term survival and prognosis of rectal cancer patients.
Footnotes
Provenance and peer review: Unsolicited article; Externally peer reviewed.
Peer-review model: Single blind
Specialty type: Gastroenterology and hepatology
Country of origin: China
Peer-review report’s classification
Scientific Quality: Grade C, Grade C
Novelty: Grade B, Grade C
Creativity or Innovation: Grade B, Grade C
Scientific Significance: Grade B, Grade C
P-Reviewer: Asik M; Liu G S-Editor: Fan M L-Editor: A P-Editor: Zhao YQ
Tan EK, Ooi LLPJ. Colorectal Cancer Liver Metastases – Understanding the Differences in the Management of Synchronous and Metachronous Disease.Ann Acad Med Singap. 2010;39:719-733.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Hołówko W, Grąt M, Hinderer B, Orlińska I, Krawczyk M. Prediction of survival in patients with unresectable colorectal liver metastases.Pol Przegl Chir. 2014;86:319-324.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Li ZF, Kang LQ, Liu FH, Zhao M, Guo SY, Lu S, Quan S. Radiomics based on preoperative rectal cancer MRI to predict the metachronous liver metastasis.Abdom Radiol (NY). 2023;48:833-843.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Lu Z, Sun J, Wang M, Jiang H, Chen G, Zhang W. A nomogram prediction model based on clinicopathological combined radiological features for metachronous liver metastasis of colorectal cancer.J Cancer. 2024;15:916-925.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Liu Y, Wang Y, Zhang H, Zheng M, Wang C, Hu Z, Wang Y, Xiong H, Hu H, Tang Q, Wang G. Nomogram for predicting occurrence of synchronous liver metastasis in colorectal cancer: a single-center retrospective study based on pathological factors.World J Surg Oncol. 2022;20:39.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Díez-Alonso M, Mendoza-Moreno F, Jiménez-Álvarez L, Nuñez O, Blazquez-Martín A, Sanchez-Gollarte A, Matías-García B, Molina R, San-Juan A, Gutierrez-Calvo A. Prognostic factors of survival in stage IV colorectal cancer with synchronous liver metastasis: Negative effect of the KRAS mutation.Mol Clin Oncol. 2021;14:93.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 1][Cited by in F6Publishing: 3][Article Influence: 1.0][Reference Citation Analysis (0)]
Puijk RS, Ruarus AH, Vroomen LGPH, van Tilborg AAJM, Scheffer HJ, Nielsen K, de Jong MC, de Vries JJJ, Zonderhuis BM, Eker HH, Kazemier G, Verheul H, van der Meijs BB, van Dam L, Sorgedrager N, Coupé VMH, van den Tol PMP, Meijerink MR; COLLISION Trial Group. Colorectal liver metastases: surgery versus thermal ablation (COLLISION) - a phase III single-blind prospective randomized controlled trial.BMC Cancer. 2018;18:821.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 71][Cited by in F6Publishing: 143][Article Influence: 23.8][Reference Citation Analysis (0)]
Adam R, de Gramont A, Figueras J, Kokudo N, Kunstlinger F, Loyer E, Poston G, Rougier P, Rubbia-Brandt L, Sobrero A, Teh C, Tejpar S, Van Cutsem E, Vauthey JN, Påhlman L; of the EGOSLIM (Expert Group on OncoSurgery management of LIver Metastases) group. Managing synchronous liver metastases from colorectal cancer: a multidisciplinary international consensus.Cancer Treat Rev. 2015;41:729-741.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 287][Cited by in F6Publishing: 372][Article Influence: 41.3][Reference Citation Analysis (1)]
Patel I, Bartlett D, Dasari BV, Chatzizacharias N, Isaac J, Marudanayagam R, Mirza DF, Roberts JK, Sutcliffe RP. Detection of Colorectal Liver Metastases Using Near-Infrared Fluorescence Imaging During Hepatectomy: Prospective Single Centre UK Study.J Gastrointest Cancer. 2023;54:574-579.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 2][Reference Citation Analysis (0)]
Granata V, Fusco R, Setola SV, Galdiero R, Maggialetti N, Patrone R, Ottaiano A, Nasti G, Silvestro L, Cassata A, Grassi F, Avallone A, Izzo F, Petrillo A. Colorectal liver metastases patients prognostic assessment: prospects and limits of radiomics and radiogenomics.Infect Agent Cancer. 2023;18:18.
[PubMed] [DOI][Cited in This Article: ][Cited by in F6Publishing: 4][Reference Citation Analysis (0)]
Han Z, Tong Y, Zhu X, Sun D, Jia N, Feng Y, Yan K, Wei Y, He J, Ju H. Development and external validation of MRI-based RAS mutation status prediction model for liver metastases of colorectal cancer.J Surg Oncol. 2024;129:556-567.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Granata V, Fusco R, Brunese MC, Ferrara G, Tatangelo F, Ottaiano A, Avallone A, Miele V, Normanno N, Izzo F, Petrillo A. Machine Learning and Radiomics Analysis for Tumor Budding Prediction in Colorectal Liver Metastases Magnetic Resonance Imaging Assessment.Diagnostics (Basel). 2024;14.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 1][Reference Citation Analysis (0)]
