CT-based radiomics nomograms for preoperative prediction of diffuse-type and signet ring cell gastric cancer: a multicenter development and validation cohort

The prevalence of diffuse-type gastric cancer (GC), especially signet ring cell carcinoma (SRCC), has shown an upward trend in the past decades. This study aimed to develop computed tomography (CT) based radiomics nomograms to distinguish diffuse-type and SRCC GC preoperatively. A total of 693 GC patients from two centers were retrospectively analyzed and divided into training, internal validation and external validation cohorts. Radiomics features were extracted from CT images, and the Lauren radiomics model was established with a support vector machine (SVM) classifier to identify diffuse-type GC. The Lauren radiomics nomogram integrating radiomics features score (Rad-score) and clinicopathological characteristics were developed and evaluated regarding prediction ability. Further, the SRCC radiomics nomogram designed to identify SRCC from diffuse-type GC was developed and evaluated following the same procedures. Multivariate analysis revealed that Rad-scores was significantly associated with diffuse-type GC and SRCC (p < 0.001). The Lauren radiomics nomogram showed promising prediction performance with an area under the curve (AUC) of 0.895 (95%CI, 0.957–0.932), 0.841 (95%CI, 0.781–0.901) and 0.893 (95%CI, 0.831–0.955) in each cohort. The SRCC radiomics nomogram also showed good discrimination, with AUC of 0.905 (95%CI,0.866–0.944), 0.845 (95%CI, 0.775–0.915) and 0.918 (95%CI, 0.842–0.994) in each cohort. The radiomics nomograms showed great model fitness and clinical usefulness by calibration curve and decision curve analysis. Our CT-based radiomics nomograms had the ability to identify the diffuse-type and SRCC GC, providing a non-invasive, efficient and preoperative diagnosis method. They may help guide preoperative clinical decision-making and benefit GC patients in the future.


Introduction
Gastric cancer (GC) is the fifth most common cancer and the third leading cause of cancer-related death worldwide [1]. Although the overall incidence of GC has significantly decreased over recent decades, the incidence of Lauren diffuse-type GC is constantly rising, and the predominant increase occurred in the

Patients
This retrospective study was approved by the institutional review board of two medical centers, and the need for informed patient consent was waived. All procedures performed involving human participants were following the 1964 Helsinki Declaration and its later amendments. Patients who underwent total or partial radical gastrectomy and histologically confirmed GC between December 2007 and March 2016 were enrolled. The detailed inclusion criteria were as follows: (1) patients who underwent surgery for GC; (2) patients who underwent standard contrast-enhanced CT less than 15 days before surgery; (3) patients with complete clinicopathologic data. Patients who received neoadjuvant chemotherapy (NAC) therapy or radiotherapy before surgery were excluded to avoid the influence of these factors on the tumor size and degree of invasion. The demographic and clinicopathologic data of patients, including age, sex, tumor site, tumor size (maximum diameter), CEA, CA199, Lauren classification, Borrmann classification, differentiation and tumor stage, were obtained from medical records. Tumor staging was performed based on the American Joint Committee on Cancer tumor-nodemetastasis (TNM) Staging Manual, 8th Edition.
Flow diagrams for eligible patients were shown in Additional file 1: Figure S1. Finally, a total of 693 patients (453 males and 240 females; mean age, 56.38 ± 11.85 years; age range, 22-87 years) from 2 medical centers were enrolled in the study, including 587 patients from center 1 (Nanfang Hospital of Southern Medical University, Guangzhou, China) and 106 patients from center 2 (Zhujiang Hospital, Guangzhou, China). To develop a Lauren radiomics model to identify diffuse-type GC, we divided all patients into three cohorts: one training cohort (n = 300 from center 1), one internal validation cohort (n = 287 from center 1) and one external validation cohort (n = 106 from center 2) (Additional file 1: Figure  S1a). Moreover, the SRCC radiomics model was designed to identify SRCC from diffuse-type GC. A total of 443 diffuse-type GC patients were included and divided into three cohorts: one training cohort (n = 280 from center 1), one internal validation cohort (n = 114 from center 1) and one external validation cohort (n = 49 from center 2) (Additional file 1: Figure S1b). The sample size consideration was shown in Additional file 1: S1.

CT image acquisition and radiomics feature extraction
The procedures of CT image acquisition and retrieval were described in detail in Additional file 1: S2. Then CT images were exported to the ITK-SNAP 3.6 (ITK-SNAP 3.X TEAM) software, and three-dimensional (3D) segmentation of the region of interest (ROI) was performed (Additional file 1: Figure S2). The algorithms for tumor ROIs delineation and reproducibility evaluation of intraobserver (reader 1 twice) and interobserver (reader 1 vs. reader 2) were described in Additional file 1: S3. The preprocessing was applied to the ROI images with different parameters (Additional file 1: Table S1) to enrich features before extracting the texture features (Additional file 1: S4). Then we applied the feature extraction method to the ROI in MATLAB 2016b (Mathworks), and series of texture features were generated from the images (Additional file 1: Table S2). Then the feature values were preprocessed with a filtering feature selection method (Additional file 1: S5).

Feature selection, construction and evaluation of the radiomics SVM models
We used the Relief forward selection (RFS) algorithm [25] and an exhaustive test based on the performance of the SVM classifier (Additional file 1: S6) to find feature subset with the best distinguishing characteristics for the radiomics model [25]. The SVM model was based on the LIBSVM software package developed by Professor Lin et al. in 2001 (https:// www. csie. ntu. edu. tw/ ~cjlin). A high penalty parameter c could effectively improve the model's accuracy, but an excessive high penalty parameter would cause over-fitting status. The range of c was limited to prevent this situation, and the tenfold cross-validation and grid search method was applied to find the best combination of SVM model parameters (c and g) (Additional file 1: Figure S3). Then all extracted features were ranked from the most important to the least important, and different feature sets were obtained using the exhaustive test from the ordered sequence 1 ≤ m ≤ M. The set of first m features was fed into the SVM classifier. Its performance for differentiating different GC types was evaluated by receiver operating characteristic (ROC) curves and the area under the curve (AUC). The detailed steps of feature selection were shown in Additional file 1: S7 in the Supplement. Differences in the AUC values between the three cohorts were assessed using the Delong test. The pathological classification radiomics feature score obtained in SVM models of each patient was seen as Rad-score. Kaplan-Meier survival analyses were used to estimate the difference in 5-year disease-free survival (DFS) and 5-year overall survival (OS) between the high Rad-score and low Rad-score groups.

Development and evaluation of radiomics nomograms
Multivariate logistic regression was applied to select independent predictors of diffuse-type and SRCC GC from the clinical characteristics. The significant predictors among the clinical characteristics and the Radscore were entered into the logistic regression analysis to develop the radiomics nomogram. The diagnostic performance and calibration of the radiomics nomogram were evaluated based on ROC and calibration curves. Decision curve analysis was applied to assess the clinical usefulness of the radiomics nomograms by quantifying the net benefit at different threshold probabilities.

Statistical analysis
Analyses were performed using SPSS version 26.0 (SPSS Inc., Chicago, IL, USA). Continuous variables were presented as the mean ± standard deviation and compared with the t-test. Categorical variables were expressed as frequency (percentage) and compared with Chi-squared tests or Fisher's exact test as appropriate. Nomograms and calibration curves were generated with the rms package of R software (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria). A p-value of < 0.05 was set as the threshold for statistical significance.

Clinical characteristics of all patients
A total of 693 patients (453 males and 240 females; mean age, 56.38 ± 11.85 years; age range, 22-87 years) were included in the study. The clinicopathologic characteristics of the assessed patients were listed in Table 1. Clinical characteristics, including tumor location, differentiation status, Borrmann type, levels of CEA and CA199, and TNM stages were significantly different between intestinal-type and diffuse-type GC patients.

Feature selection and construction of the Lauren radiomics SVM model
A total of 9691 features were extracted from the tumor ROI with satisfactory interobserver and intraobserver reproducibility assessments (Additional file 1: S8). The weight ordering of radiomics features was obtained by the Relief algorithm (Fig. 1a). The feature subset with the best discrimination ability for the radiomics model was obtained using the exhaustive test based on the performance of the support vector machine (SVM) classifier. Finally, the optimal feature subset with 13 features achieved excellent performance in distinguishing Lauren diffuse-type and intestinal-type GC ( Fig. 1b and Additional file 1: Table S3), yielding AUC values of 0.895 (95% confidence interval (CI) 0.957-0.932), 0.791 (95%CI0.728-0.853) and 0.857 (95%CI0.78-0.935) in the training, internal validation and external validation cohort, respectively (Fig. 1c). Multivariate analysis revealed that Rad-score was the significant predictor between intestinal-type and diffuse-type GC (OR, 4.164; 95%CI, (3.121,5.557); p < 0.001) (Additional file 1: Table S4). Further, statistical difference was found in terms of 5-year DFS and OS between the high Rad-score group (diffuse-type) and low Rad-score group (intestinal-type) (Additional file 1: S9).  Table 2). The Rad-score and clinical characteristics were incorporated to build the Lauren radiomics nomogram (Fig. 2). The diagnostic performance comparison of the Lauren radiomics model and radiomics nomogram was shown in Fig. 3. No difference was observed in the training cohort (Fig. 3a), while the radiomics nomogram achieved a higher AUC than the radiomics model in the internal validation cohort ( . 3b and 3c). The Lauren radiomics nomogram model had higher specificity, sensitivity and accuracy than the SVM model (Additional file 1: Table S5). The Delong test was applied on the ROC curves of the radiomics nomogram to assess possible overfitting and the result revealed that the differences were not statistically significant among the AUCs of the training cohort and the two validation cohorts, with P values of 0.138 and 0.969, respectively. The calibration curves demonstrated good agreement between prediction and observation in all three cohorts (Hosmer-Lemeshow test, p > 0.05) ( Fig. 3d-f ). The decision curve analysis (Additional file 1: Figure S6) indicated that the patients would benefit more from using the radiomics nomograms than using the SVM model or treat-all-patients scheme or the treatnone scheme if the threshold probability in the clinical decision was between 10 and 90%.

Clinical characteristics of patients with diffuse-type GC
To further develop the SRCC radiomics model to distinguish SRCC and non-SRCC in diffuse-type GC patients, 394 diffuse-type GC patients from center 1 and 49 patients from center 2 were enrolled. The clinicopathologic characteristics of these patients were listed in Table 3. Clinical characteristics, including tumor location, differentiation status, Borrmann type, levels of CEA and CA199, and TNM stages were significantly different between non-SRCC and SRCC GC patients.

Construction and evaluation of the SRCC radiomics nomogram
The diagnostic performance comparison of the SRCC radiomics model and radiomics nomogram was shown in Fig. 6. No obvious differences were observed in the training cohort (Fig. 6a), while the radiomics nomogram achieved higher AUC than the radiomics model in the internal validation cohort (AUC, 0.845 [95%CI 0.775-0.915] vs 0.824 [95%CI 0.748-0.900]) (Fig. 6b) (Fig. 6c). The SRCC radiomics nomogram model had higher specificity, sensitivity and accuracy than the SVM model (Additional file 1: Table S7). The Delong test revealed that the differences were not statistically significant among the AUCs of the training cohort and the two validation cohorts, with P values of 0.138 and 0.969, indicating no overfitting was assessed. The calibration curves of the radiomics nomogram demonstrated good agreement between prediction and observation in all three cohorts (Hosmer-Lemeshow test, p > 0.05) (Fig. 6d-f ). The decision curve analysis indicated that the patients would benefit more from using the radiomics nomograms than using the SVM model or treat-all-patients scheme or the treat-none scheme if the threshold probability in the clinical decision was between 10 and 90% (Additional file 1: S9).

Discussion
In this retrospective multicenter study, we established a CT-based Lauren radiomics nomogram to identify the diffuse-type GC from all GC patients and further developed a SRCC radiomics nomogram to identify SRCC from diffuse-type GC. The nomograms provided a noninvasive and efficient preoperative diagnosis method to identify diffuse-type and SRCC GC.
Lauren classification is one of the most widely used histopathological classification systems for gastric adenocarcinoma [5,26]. In addition to reflecting   tumor biological behavior, it can also reflect the etiology, pathogenesis and epidemic characteristics of GC. Diffuse-type GC, which originates from the gastric mucosa and exhibits a diffuse growth pattern, is poorly differentiated and shows more chemotherapy resistance [27]. It is more prone to lymph node metastasis and distant metastasis than intestinal-type, resulting in a poor prognosis [28]. Studies have found that germline mutations in some genes (such as CDH1, BRCA2, STK11, ATM and PALB2) may be the cause of diffuse-type GC [14,29,30]. According to epidemiological data, there has been an increasing trend in the incidence of diffuse-type GC [3]. As a result, the early diagnosis and treatment of diffuse-type GC have attracted widespread attention worldwide. Gastroscopy and tissue biopsy are the most commonly used methods for the pathological diagnosis of GC. However, they are invasive operations, and the consistent rate of the Lauren classification was only 64.7% between biopsy and surgical samples [16]. The recent emergence of radiomics undoubtedly provides an excellent solution to this problem. In this study, 693 GC patients from 2 centers were retrospectively analyzed, and 9691 radiomics features were extracted from their CT image. Radiomics feature subset with the best distinguishing characteristics was searched by SVM classifier to develop the Lauren radiomics model. SVM is a mature machine learning method with relatively stable performance and gradually replaces the previous lasso regression method. Multivariate analyses revealed that radiomics feature score could be the independent predictor of diffuse-type GC. Then, the Lauren radiomics nomogram integrating Rad-score and clinicopathological characteristics was developed, which was proved a promising AUC value and satisfactory calibration. Age, tumor size and CEA levels were found significantly associated with diffuse-type GC in this study, consistent with our previous literature review [16,31].
SRCC, as a particular type of diffuse-type GC, is characterized by a higher incidence in females and a lower  average age at diagnosis than non-SRCC [32]. Meanwhile, it has a higher rate of peritoneal carcinomatosis, lymph node invasion and chemotherapy resistance and a lower curative resection rate than non-SRCC tumors in advanced stages [2,9,14,33]. Moreover, SRCCs often manifest as Borrmann IV type with a high false-negative rate during biopsy [17]. Considering the importance of early diagnosis, we further develop another radiomics SVM model (SRCC radiomics model) to identifying SRCC from diffuse-type GC. Multivariate analyses revealed that the model's Rad-score could be the independent predictor of SRCC. Further, the SRCC nomogram integrating Radscore and clinicopathological characteristics including sex and tumor location was developed. The results showed that the SRCC radiomics nomogram had higher AUC values and accuracy than the radiomics SVM model and the decision curve analysis demonstrated that the radiomics nomogram was clinically valuable. In addition, nomograms in this study may also help future clinical decision making. Different pathological types of GC have different benefits from the same treatment, so it is necessary to choose appropriate treatment measures according to pathological types. For example, as for surgical management, diffuse-type GC usually need wider surgical margins to achieve an R0 resection, and a super-extended lymphadenectomy might be the best surgical approach [34]. A survival benefit with D3 lymphadenectomy, compared with D2 lymphadenectomy, can be obtained in diffuse-type and mixed-type GC [35]. In addition, diffuse-type GC may benefit from prevention and/or treatment of peritoneal metastases using hyperthermic intraperitoneal chemotherapy (HIPEC) [34,36]. Therefore, if diffuse-type GC can be diagnosed and distinguished in an early stage, it will be of great help to the choice of treatment schemes and the prognosis evaluation.
There are some limitations to our study. First, as it was a retrospective study involving only two centers, further prospective research in more centers is needed to verify the radiomics nomograms. Second, SRCC is a special histological type with different clinical outcomes, depending on whether it is in an early or advanced stage [12,18]. However, in this study, we did not perform analysis on this issue, only focused on the diagnosis of SRCC. Further radiomics research with subgroup analysis should be performed to reveal more biological characteristics of SRCC.

Conclusion
In summary, we established two CT-based radiomics nomograms to identify the diffuse-type and SRCC GC, providing a noninvasive, efficient and preoperative diagnosis method. They may help guide preoperative clinical decision-making and benefit GC patients in the future.