Combining Clinicopathology, IVIM-DWI and Texture Parameters for a Nomogram to Predict Treatment Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer Patients

Objectives This study aimed to create a nomogram for the risk prediction of neoadjuvant chemoradiotherapy (nCRT) resistance in locally advanced rectal cancer (LARC). Methods Clinical data in this retrospective study were collected from a total of 135 LARC patients admitted to our hospital from June 2016 to December 2020. After screening by inclusion and exclusion criteria, 62 patients were included in the study. Texture analysis (TA) was performed on T2WI and DWI images. Patients were divided into response group (CR+PR) and no-response group (SD+PD) according to efficacy assessment. Multivariate analysis was performed on clinicopathology, IVIM-DWI and texture parameters for screening of independent predictors. A nomogram was created and model fit and clinical net benefit were assessed. Results Multivariate analysis of clinicopathology parameters showed that the differentiation and T stage were independent predictors (OR values were 14.516 and 11.589, resp.; P<0.05). Multivariate analysis of IVIM-DWI and texture parameters showed that f value and Rads-score were independent predictors (OR values were 0.855, 2.790, resp.; P<0.05). In this study, clinicopathology together with IVIM-DWI and texture parameters showed the best predictive efficacy (AUC=0.979). The nomogram showed good predictive performance and stability in identifying high-risk LARC patients who are resistant to nCRT (C-index=0.979). Decision curve analyses showed that the nomogram had the best clinical net benefit. Ten-fold cross-validation results showed that the average AUC value was 0.967, and the average C-index was 0.966. Conclusions The nomogram combining the differentiation, T stage, f value and Rads-score can effectively estimate the risk of nCRT resistance in patients with LARC.


INTRODUCTION
Rectal cancer is the third most common malignancy in terms of morbidity and mortality worldwide, and nearly half of rectal cancers are diagnosed as locally advanced rectal cancer (LARC) (1). Total mesorectal excision (TME) after neoadjuvant chemoradiotherapy (nCRT) for LARC is the current standard treatment modality, which can improve the local control rate, sphincter preservation rate, and reduce the local recurrence rate of LARC. And 10% to 20% of patients can achieve pathological complete remission after nCRT (2). Early assessment of nCRT efficacy and preservation of anal sphincter function can significantly improve the quality of life for patients (3,4). Therefore, it becomes an urgent issue to find markers that can predict the efficacy of nCRT for LARC, then differentiate patients who are sensitive or resistant to nCRT at an early stage, and formulate individualized treatment for different patients.
Since the introduction of conventional MRI to local staging of rectal cancer, many efforts have been made to find predictive imaging signatures so as to identify patients with good or poor response to nCRT and patients at higher risk of recurrence (5,6). However, as conventional MRI has poor sensitivity and specificity in assessing and predicting treatment response, it is hard to distinguish post-treatment fibrosis and edema from tumor tissue after nCRT, and even harder to accurately predict complete pathological remission (7)(8)(9). Without the use of exogenous contrast agents, intravoxel incoherent motion diffusion weighted imaging (IVIM-DWI) based on the biexponential model can assess the pure diffusion motion and perfusion-related motion of water molecules separately. Compared to conventional MRI, IVIM-DWI can more accurately display tissue microenvironment information and non-invasively and quantitatively diagnose tumor malignancy, pathological differentiation, and lymph node metastasis (10)(11)(12). In recent years, several studies have proved the superiority of IVIM-DWI in predicting treatment response in various tumors (13)(14)(15). It was found that IVIM-DWI was effective in predicting tumor size changes in patients with breast cancer liver metastases who were undergoing radioembolization, and that treatmentinduced changes in f value could be a potential biomarker in the prediction of treatment response (16). Studies have also shown that IVIM-DWI parameters, especially pretreatment D value could predict patients' response to induction chemotherapy with locally advanced hypopharyngeal carcinoma (17). However, no conclusive results have been obtained about the efficacy of IVIM-DWI in predicting the response of patients with LARC to nCRT.
Texture analysis (TA) is an emerging image analysis method that extracts quantitative imaging features from images through high throughput analysis. It can reflect the spatial variation and heterogeneity of voxel intensities within the tumor and can provide valuable references for disease diagnosis, treatment, and efficacy prediction of therapies (18,19), thus a promising method for assessing response to various cancer treatments. Studies have found that texture analysis of magnetic resonance imaging (MRI) could predict early recurrence after hepatectomy for solitary hepatocellular carcinoma (20). It was also found that MRI texture features could predict pathological complete response (pCR) to neoadjuvant chemotherapy in breast cancer (21). In rectal cancer, TA can work as a supplement to conventional MR imaging, especially in identifying staging, assessing treatment response, and predicting lymph node metastasis and prognosis (22)(23)(24).
The purpose of this study was to combine clinicopathology, IVIM-DWI and texture parameters to establish a nomogram for treatment response prediction to nCRT in patients with LARC, which can be used to identify patients at high risk of nCRT resistance.

Patients
A total of 135 LARC patients admitted to Anhui Provincial Hospital from June 2016 to December 2020 were retrospectively collected, among whom 62 patients with complete clinical data were included in this study. Inclusion criteria included (1) Locally advanced rectal primary adenocarcinoma was confirmed by colonoscopy biopsy; (2) All patients received pelvic nCRT before surgery; (3) The lower edge of the tumor is within 15 cm of the anal verge; (4) Patients completed routine MRI and IVIM-DWI examinations before nCRT. Exclusion criteria were (1) Patients who have not received complete nCRT; (2) Patients with a history of other tumors; (3) The image quality is poor, with significant artifacts, and cannot be used for image segmentation or image histology feature extraction. The flow chart of this study is shown in Figure 1.

IVIM-DWI Measurement
The IVIM-DWI measurement was performed by two radiologists. They used FunctionTool to build regions of interest (ROI) on a post-processing workstation (version ADW 4.5, GE Healthcare) without knowledge of clinicopathological results. To acquire the parameters, the ROI of IVIM-DWI was selected on the maximum transverse plane of each lesion (b value=800 s/mm 2 ) ( Figure 2). Referring to the axial T2WI image, the ROI was manually delineated along the tumor margin, avoiding the hemorrhage, necrosis, and cystic change of the tumor. All parameters were measured three times and the average value was calculated. To assess interobserver agreement, each radiologist plotted twice to obtain values and calculate intraclass correlation coefficients (ICC). Parameters with ICC value > 0.75 were selected for subsequent analysis.

Texture Features Analyses
Patients underwent tumor delineation and texture features extraction. The raw data was transferred to the PACS system, and axis IVIM-DWI (b=800 s/mm 2 ) and T 2 WI images (DICOM    Figure 3 shows the texture feature extraction.
Features with ICC value > 0.75 were selected for subsequent analysis.
nCRT Regimen nCRT Efficacy Evaluation nCRT efficacy on patients was assessed 6-8 weeks after they received nCRT. Two senior physicians with rich experience in

Statistical Analysis
All data were statistically analyzed using Graphpad Prism 9.0, SPSS 26.0, R 4.0.5 and IPMS 2.4.0 software. For continuous variables, Kolmogorov-Smirnov was used to test the normality of the data. If the data followed normal distribution, x ± s was used for statistical description, and independent sample t-test was conducted to compare two groups; otherwise, median ± interquartile range was used for statistical description, and Mann-Whitney U test was conducted to compare two groups. For categorical variables, the number and percentage of cases were used for statistical description, and the chi-square test or Fisher's exact probability was adopted for comparison between two groups. ICC assessed the inter-observer and intra-observer agreement of IVIM-DWI values and texture feature measurements (ICC>0.75 indicated good agreement) (26,27).
Radiomic features were screened by independent samples t-test or Mann-Whitney U test and multivariate logistic regression. Each set of texture features was developed using multi-factor linear weighting based on the regression coefficients of the selected features, and a radiomics score (Rads-score) was calculated for each patient. A nomogram model was established on the basis of clinicopathology, IVIM-DWI and Rads-score parameters. Ten-fold cross-validation was performed to assess discrimination and prediction ability of the nomogram model. All statistical tests were two-sided probability tests, and P <0.05 was considered statistically significant.

Ethical Approval and Informed Consent
This study was approved by the Ethics Committee of Anhui Provincial Hospital and the requirement for informed consent was waived due to the retrospective nature of this study.

Clinicopathology Characteristics
A total of 62 patients' data were collected. The average age of the patients was 58.77 ± 15.66 years. The clinical characteristics of the patients were shown in Table 2. All patients completed the efficacy evaluation, including 6 cases of CR (9.68%), 38 cases of PR (61.29%), 18 cases of SD (29.03%), 0 cases of PD (0%); ORR was 70.97% and DCR was 100%. There were no significant differences in age, gender, tumor diameter, N stage, CA199, and HB between the two groups (P>0.05).

Intra-and Inter-Observer Agreement
Among the 1 656 texture features extracted from T2WI and IVIM-DWI (b=800 s/mm 2 ) images, there were 1 054 text features with were independent factors affecting nCRT efficacy ( Table 3). Therefore, a predictive model (model 1) was developed.

Predictive Performance of IVIM-DWI and Texture Parameters
Univariate analysis showed that D, D*, f and Rads-score were significantly correlated with the efficacy of nCRT in patients (P<0.05). Multivariate logistic analysis showed that f (OR: 0.855, 95%CI: 0.752-0.971, P=0.016) and Rads-score (OR: 2.790, 95% CI: 1.163-6.696, P=0.022) were independent factors influencing nCRT efficacy in patients (Table 4). Therefore, a predictive model (model 2) was built to describe the relations between independent factors and nCRT efficacy.

Development and Performance of the Nomogram
Based on combination of the differentiation, T stage, f value and Rads-score, a nomogram was created to predict tumor resistance to nCRT in patients with LARC ( Figure 4). AUCs of differentiation, T stage, differentiation and T stage, f, Radsscore, f and Rads-score and the nomogram were 0.633, 0.707, 0.778, 0.857, 0.783, 0.917, 0.979, respectively ( Figure 5) ( Table 5). The nomogram calibration curve revealed that the observation and prediction were conducted in a good agreement ( Figure 6A). C-index was 0.979, and Brier score was 0.052. Decision curve analysis (DCA) showed that the model developed by combining the differentiation, T stage, f value and Rads-score had the largest net benefit ( Figure 6B).

Internal Validation of the Nomogram Model
To assess model's discrimination and prediction ability, the nomogram model was internally validated using ten-fold crossvalidation. The results showed that the average AUC value was 0.967, and the average C-index was 0.966.

DISCUSSION
In this study, it was found that the differentiation, T stage, f value and Rads-score were related to tumor resistance to nCRT in patients with LARC. We developed and validated a nomogram which incorporated the differentiation, T stage, f value and Rads-score. The results showed that the nomogram could yield the highest AUC. Calibration curve of the nomogram revealed that the observation and prediction were in a good agreement. DCA indicated that the nomogram model had the largest net benefit.
Since the introduction of quantitative imaging, an increasing number of studies on quantitative imaging have been conducted to improve the diagnosis and treatment of cancer and many of these studies had demonstrated the widespread use of quantitative imaging in the clinical setting (28,29). For instance, Ren et al. established an MRI-based radiomic signature from combined contrast-enhanced T1-weighted and T2-weighted images to predict the preoperative staging of head and neck cancers (30). Zheng et al. combined the IVIM-DWI  (26). In this study, we combined the differentiation, T stage, f value and Rads-score to establish a nomogram and our nomogram showed good performance in predicting treatment response to nCRT in LARC patients (AUC=0.967). This study found that poor differentiation and high T stage were associated with poor response to nCRT in patients with LARC, and these two factors were also independent predictors of tumor resistance to nCRT. Therefore, we established a predictive model (model 1) which combined the differentiation and T stage.
The AUC of the model was 0.778, suggesting that the model had a certain predictive value. In a previous study, Huang et al. found that patients in the poorly differentiated group had a worse prognosis than the non-poorly differentiated group who received nCRT for rectal cancer treatment (32). Xu et al. found that in T2N0M0 colorectal cancer patients, the well-differentiated group had a better prognosis than the poorly differentiated group, and the differentiation was an independent prognostic factor in T2N0M0 colorectal cancer (33). The poor differentiation has been shown to be associated with bowel penetration, lymph node involvement and vascular invasion, suggesting that it is a risk factor for colorectal cancer invasion and dissemination (34). T stage reflects the depth of tumor infiltration and correlates with cancer invasion, and several studies have explored the prognostic value of T stage in the colorectum (35)(36)(37). It was found that the T4 stage was an independent risk factor for tumor resistance to nCRT in LARC. In a retrospective study, Xu et al. found that the T4 stage was an independent risk factor for shorter overall survival (OS) in patients with colorectal cancer   (39). In summary, the poor differentiation and high T stage of rectal cancer may reflect the high degree of malignancy, which may lead to tumor resistance to nCRT and poor efficacy.
In addition to clinicopathological parameters, this study also found that low f value and high Rads-score were significantly associated with poor efficacy. We combined f and Rads-score to establish the second prediction model (model 2) of tumor resistance to nCRT in LARC and the model had a good predictive performance, AUC=0.917. The f value represents the voxel's blood volume fraction, which can reflect the capillary density and is related to the blood perfusion in the tissue (40).    (43). These studies confirmed that LARC patients with high pretreatment f value had a better response to nCRT. TA is a quantitative image post-processing technique, which can objectively reflect the underlying biological characteristics and heterogeneity of tumors due to its quantitative extraction and analysis of the pixel distribution of the lesion area (44). In a previous study, Lu et al. found that High DISS (gray-level runlength matrix_Dissimilarity) on sagittal fat-suppression T2WI and high DISS and ENTR (gray-level co-occurrence matrix_Entropy) on transverse T2WI could predict high T stage in rectal cancer (45). In a study of TA for early prediction of patients' response to nCRT in LARC, Park et al. found that GLRLM_LRLGE (gray-level run-length matrix_Long Run Low Gray Level Emphasis) on T2WI images could predict tumor recurrence after treatment (46). In this study, the texture parameters of DWI and T2WI images were analyzed and we filter out these two parameters [glrlm_LongRunEmphasis (T2WI) and firstorder_Median (IVIM)] through texture features and calculated the Rads-score, both of which were found to be positively correlated with Rads-score. Multivariate results showed that Rads-score was an independent risk factor for tumor resistance to nCRT in LARC. The glrlm_LongRun Emphasis is a measure of long run length distribution, with a higher value indicating longer run lengths and coarser structural textures. The firstorder_Median is the ROI's median gray level intensity. These two parameters can represent heterogeneity and subtle changes within the tumor. The results of this study showed that the rough texture and gray intensity in the RIO were more resistant to nCRT.
Most previous studies on rectal cancer only explored the role of IVIM-DWI or texture parameters of a single sequence image (47)(48)(49). In this study, two sequence image texture parameters were analyzed and combined with clinicopathology, IVIM-DWI and texture parameters to establish a nomogram for predicting tumor resistance to nCRT in LARC, thus richer and more accurate than the results obtained in previous studies. Compared with model 1 and model 2, the nomogram had higher predictive power and larger net benefit. The nomogram established a risk prediction system for tumor resistance to nCRT in LARC, so that appropriate treatment can be formulated for different patients, which is beneficial to patient recovery.
There are several limitations of this study. First, the manual image segmentation process might be affected by some subjective and objective factors. Second, some cases with small lesions were excluded due to the need for sufficient pixels to ensure the reliability of IVIM-DWI and TA parameter measurements, therefore, the results we obtained might be biased. Third, this study lacks validation with an external cohort. Finally, the present study was a single-center retrospective study with small sample size. Thus, it is necessary to conduct a multicenter prospective study with a large sample size for verification in the future.

CONCLUSIONS
In conclusion, the differentiation, T stage, f value and Rads-score were independent predictors of tumor resistance to nCRT in LARC. The nomogram model combining the differentiation, T stage, f value and Rads-score showed promising performance in estimating the risk of tumor resistance to nCRT in patients with LARC.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Anhui Provincial Hospital Ethics Committee. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.