Efficacy and prognostic value of delta radiomics on dual-energy computed tomography for gastric cancer with neoadjuvant chemotherapy: a preliminary study

Abstract

Background

A non-invasive tool for tumor regression grade (TRG) evaluation is urgently needed for gastric cancer (GC) treated with neoadjuvant chemotherapy (NAC).

Purpose

To develop and validate a radiomics signature (RS) to evaluate TRG for locally advanced GC after NAC and assess its prognostic value.

Material and Methods

A total of 103 patients with GC treated with NAC were retrospectively recruited from April 2018 to December 2019 and were randomly allocated into a training cohort (n = 69) and a validation cohort (n = 34). Delineation was performed on both mixed and iodine-uptake images based on dual-energy computed tomography (DECT). A total of 4094 radiomics features were extracted from the pre-NAC, post-NAC, and delta feature sets. Spearman correlation and the least absolute shrinkage and selection operator were used for dimensionality reduction. Multivariable logistic regression was used for TRG evaluation and generated the optimal RS. Kaplan–Meier survival analysis with the log-rank test was implemented in an independent cohort of 40 patients to validate the prognostic value of the optimal RS.

Results

Three, five, and six radiomics features were finally selected for the pre-NAC, post-NAC, and delta feature sets. The delta model demonstrated the best performance in assessing TRG in both the training and the validation cohorts (AUCs=0.91 and 0.76, respectively; P>0.1). The optimal RS from the delta model showed a significant capability to predict survival in the independent cohort (P<0.05).

Conclusion

Delta radiomics based on DECT images serves as a potential biomarker for TRG evaluation and shows prognostic value for patients with GC treated with NAC.

Keywords

Stomach neoplasms X-ray computed tomography neoadjuvant treatment image processing computer-assisted

Introduction

Gastric cancer (GC) is the third leading cause of cancer fatality worldwide, accounting for approximately 783,000 deaths in 2018 (1). China is a high incidence area for GC, leading to nearly half of GC-related deaths worldwide annually (2). More than 80% of cases in China are initially diagnosed as locally advanced diseases with poor prognosis (3). Neoadjuvant chemotherapy (NAC) followed by complete surgical resection is generally accepted to improve prognosis in patients with locally advanced GC compared with resection alone, with an increasing five-year overall survival in the range of 10%–15% (4).

The therapeutic strategy mainly relies on treatment response or non-response after neoadjuvant treatment for primary tumors, which is determined by pathological assessment of subsequent surgical specimens (5). Tumor regression grading (TRG) systems are proposed to classify the degree of regressive changes after cytotoxic treatment. Evaluation of tumor regression provides potential prognostic relevance, as is reported in studies on GC (6 –8). However, at least 20% of cases present non-response after neoadjuvant treatment, leading to unnecessary cytotoxicity (9). In addition, TRG for patients is invasive and depends on pathologists’ experience. Therefore, a non-invasive and alternative tool for TRG evaluation is urgently needed.

Radiomics is an emerging field for providing insight into tumor heterogeneity by non-invasively extracting and analyzing a large number of mathematic features from medical images (10). Computed tomography (CT) is the most commonly used screening modality for GC and CT radiomics has been used to predict therapeutic response and prognosis for GC, including NAC, chemotherapy, and radical resection (9,11,12). Compared with conventional CT, dual-energy CT (DECT) is an emerging modality for stomach imaging due to its insight of pathophysiological information (13). There have been many studies making advances in DECT-based radiomics (14 –16). Zhou et al. developed a model based on DECT-derived radiomics that significantly outperformed qualitative CT image features in differentiating metastatic from non-metastatic lymph nodes of papillary thyroid cancer (14). Li et al. found that a DECT-based deep learning radiomics nomogram could predict lymph node metastasis in GC (15). Moreover, our team also found that DECT-derived radiomics could serve as a non-invasive and easy-to-use biomarker to preoperatively predict peritoneal metastasis for GC (16). However, most GC-related studies performed radiomics analysis based on pretreatment CT images alone, which is perhaps insufficient to inflect benefits from radiomics (17). Few studies have evaluated the relationship between TRG and prognosis after NAC using radiomics for locally advanced GC (18–19).

The aim of the present study was to develop and validate an optimal radiomics signature based on pre- and post-NAC CT images to evaluate TRG for patients with locally advanced GC. The secondary aim was to apply the optimal radiomics signature to a Cox model to predict the prognosis for followed-up patients with GC in an independent cohort.

Material and Methods

Patients

This retrospective study was approved by the ethics committee of Ruijin Hospital (No. 2016[88]), and the requirement for informed consent was waived. All procedures involving human participants adhered to the tenets of the Declaration of Helsinki.

From April 2018 to December 2019, consecutive patients diagnosed with GC by gastroscopic biopsy were enrolled in this study. All patients received standard NAC before undergoing total or partial radical gastrectomy and TRG after NAC was evaluated. The inclusion criteria were as follows: (i) histologically confirmed gastric adenocarcinoma diagnosed as locally advanced GC (_CT3-4a/bNxM0) based on CT examination; (ii) availability of pathological results after surgery; and (iii) DECT scans performed <3 weeks before and after NAC. The exclusion criteria were as follows: (i) poor imaging quality due to movement artifacts or other reasons (n = 9); (ii) intolerance to NAC due to poor performance status (n = 15); and (iii) complicated with concurrent cancer (n = 2).

To determine whether there are any relationships between evaluation of TRG and GC prognosis, we also investigated the value of the radiomics signature to predict the prognosis of patients after NAC. Due to the 103 patients enrolled newly receiving NAC and not having adequate follow-up, we further enrolled an independent cohort of patients from July 2017 to March 2018. The inclusion and exclusion criteria for the patients were consistent with those for TRG evaluation. Accordingly, a total of 40 patients (32 men, 8 women; median age = 59.0 years; interquartile range = 53.5–64.0 years) were enrolled for further prognostic prediction as an independent cohort.

The clinical characteristics collected included sex, age, cT stage, cN stage, and tumor stage according to the 8th American Joint Committee on Cancer (AJCC) cancer stage manual before NAC, tumor differentiation, Bormann type, tumor location, and laboratory indices before and after NAC, including alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), carbohydrate antigen 199 (CA-199), carbohydrate antigen 125 (CA-125), and carbohydrate antigen 724 (CA-724). A flow chart of the present study is presented in Fig. 1. NAC protocol was implemented as described in Supplemental Material S1.

Evaluation of NAC response and prognosis

TRG after NAC was evaluated by the Becker criterion (20). There were 40 patients in the response group and 63 patients in the non-response group. Disease-free survival (DFS) and overall survival (OS) were evaluated for prognosis in the independent cohort. Information for TRG evaluation and prognosis are shown in Supplemental Material S2.

Imaging protocol

All patients underwent DECT by one of two scanners (SOMATOM Definition Flash or SOMATOM Force; Siemens Healthineers, Forchheim, Germany) after fasting overnight. The imaging protocol was described in our previous study (21) and detailed information is presented in Supplemental Material S3.

All portal venous phase and delayed phase images were reconstructed by a D30f kernel and mixed at 120 kV with a linear blending technique using a slice thickness of 1.5 mm. The images then underwent postprocessing by delivering to a dedicated workstation with dual-energy software (Syngo.via version VB10; Siemens Healthineers). Apart from the 120-kV equivalent mixed images, the iodine uptake (IU) maps were reconstructed and obtained from the dual-energy datasets (22,23). The IU map was based on the dual-energy, three-material decomposition algorithm and represented the absolute IU value in the field of view (24). These mixed images and IU DECT images based on both the portal venous phase and delayed phase were ultimately acquired for further delineation and analysis.

Tumor delineation and feature extraction

The CT images described above, including two sets (pre-NAC and post-NAC sets) for each patient, were used for segmentation. Delineation of the region of interest (ROI) and extraction of radiomics features were completed by a radiomics analysis software on a research platform (Syngo Via version VB10; Research Frontier, Siemens Healthineer). The gastric tumor was semi-automatically delineated around the border of the tumor slice by slice on CT images by two radiologists (with 10 and 5 years of experience in abdominal imaging), omitting the first and last slices to avoid partial volume effects. We performed delineation on both pre-NAC and post-NAC DECT images.

Radiomics features were based on the PyRadiomics library; all features were reproducible and matched the benchmarks of the image biomarker standardization initiative (25,26). Three categories of radiomics features were analyzed, including 18 first-order features, 75 texture features, and 17 shape features. Except for the original radiomics features extracted above, we also analyzed radiomics features customized by several image preprocessing producers, including four types of filters and eight types of wavelet transformation (Supplemental Material S4 and Supplemental Fig. S1). Finally, 1226 radiomics features were extracted in each CT set. Therefore, 4904 (1226 × 2 [mixed and IU images] × 2 [portal venous and delayed images]) radiomics features were extracted for each patient on pre-NAC or post-NAC DECT images. To determine the best model, we also performed a delta radiomics set by calculating the change rate of radiomics features (delta radiomics set = features_post–features_pre/ features_post). Therefore, there were three radiomics feature sets (each feature set contained 4904 radiomics features) applied for TRG evaluation.

Dimensionality reduction and modeling for TRG evaluation

To avoid the curse of dimensionality and filter important radiomics features, a stepwise method was adopted. First, 50 patients were randomly selected, and radiologists 1 and 2 independently performed delineation ROIs once for the 50 patients on mixed DECT images on portal venous phase pre-NAC treatment. Spearman correlation analysis was performed to filter robust radiomics features between two delineations (i.e. features with r > 0.8 were considered highly correlated and robust). Then the least absolute shrinkage and selection operator (LASSO) was applied to diminish redundancy, with the minimum criterion using 10-fold cross-validation (representing minimum mean-squared error when performing feature selection) to tune the regularization parameter (λ).

For modeling, multivariable logistic regression was performed to predict the TRG of GC after NAC in the training cohort. Three models based on different radiomics sets were constructed and compared, including the pre-NAC model, post-NAC model, and delta model. The performance of the models was further assessed and compared in the validation cohort. The radiomics signature was chosen from the best radiomics TRG model constructed by the radiomics feature set with the best discrimination capability.

Prognosis prediction for patients with GC

We further investigated the prognostic value of the generated optimal radiomics signature. The relationship between the radiomics signature and DFS and OS was assessed in the independent cohort using Kaplan–Meier survival analysis with the log-rank test with the high-risk and low-risk groups discriminating from the optimal cutoff of the radiomics signature. Cox proportional hazard model analysis was implemented to evaluate the optimal radiomics signature as an independent biomarker for both DFS and OS.

Statistics

All statistical analyses were conducted by R version 4.0.0. Clinical characteristics were measured based on different data distributions. Continuous variables were described as the means or medians and were compared using independent t tests or Wilcoxon rank-sum tests. Categorical variables were described as proportions and were analyzed by chi-square tests or Fisher's exact tests. Discrimination metrics were used for evaluation of TRG including area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Kaplan–Meier curves were analyzed by the log-rank test. Harrell's concordance index (C-index) was measured for comparison when modeling with Cox regression. Independent predictors for evaluation of TRG and prognosis prediction by odds ratio (OR) and hazard ratio (HR) and related 95% confidence interval (CI) were also reported. Feature selection for LASSO regression was performed using the glmnet package. Modeling for Cox regression was performed using the survival package. Receiver operating characteristic (ROC) curves and nomograms were plotted using the pROC and rms packages. A two-sided P < 0.05 was considered significantly different.

Results

Patients

Finally, there were 103 patients (68 men, 35 women; mean age = 59.3 ± 10.2 years) enrolled in this study and they were randomly divided into the training cohort (n = 69) and the validation cohort (n = 34) by a 2:1 ratio. There was no bias in terms of response rate with TRG evaluation in the training and validation cohort (33.8% vs. 50.0%; P = 0.173). Sex and age in the two cohorts also showed no significant difference (sex [female] = 32.4% vs. 20.6%; P = 0.099; age = 60.4 ± 9.6 years vs. 57.1 ± 11.0 years; P = 0.114). No clinical indices indicated significant difference for TRG evaluation in the training cohort or validation cohort (all P > 0.05), except for one parameter, pre-CA199, which presented significance in the validation cohort (P = 0.005). Detailed information is summarized in Table 1.

Table 1.

Clinical characteristics for evaluation of TRG in the training cohort and validation cohort.

	The training cohort		P	The validation cohort		P
	Response (n = 23)	Non-response (n = 46)	P	Response (n = 17)	Non-response (n = 17)	P
Age (years)	62.2 ± 10.0	59.5 ± 9.4	0.279	60.6 ± 10.6	53.5 ± 10.7	0.061
Sex			0.928			0.720
Male	16 (69.6)	30 (65.2)		10 (58.8)	12 (70.6)
Female	7 (30.4)	16 (34.8)		7 (41.2)	5 (29.4)
Pre_cT stage			0.722			>0.999
T3	1 (4.3)	0 (0)		0 (0)	0 (0)
T4	22 (95.7)	46 (100)		17 (100)	17 (100)
Pre_cN stage			0.140			0.197
N0	2 (8.7)	2 (4.3)		1 (5.9)	1 (5.9)
N1	1 (4.3)	10 (21.7)		7 (41.2)	3 (17.6)
N2	10 (43.5)	23 (50.0)		4 (23.5)	10 (58.8)
N3	10 (43.5)	11 (23.9)		5 (29.4)	3 (17.6)
AJCC stage			0.574			0.102
II	2 (8.7)	2 (4.3)		1 (5.9)	1 (5.9)
III	16 (69.6)	37 (80.4)		12 (70.6)	16 (94.1)
IV	5 (21.7)	7 (15.2)		4 (23.5)	0 (0)
Tumor differentiation			0.516			0.888
Poor	20 (87.0)	37 (80.4)		13 (76.5)	14 (82.4)
Moderate	1 (4.3)	6 (13.0)		3 (17.6)	2 (11.8)
SRCC	5 (21.7)	3 (6.5)		1 (5.9)	1 (5.9)
Bormann type			0.621			0.079
I	0 (0)	3 (6.5)		1 (5.9)	0 (0)
II	16 (52.2)	23 (50.0)		7 (41.2)	13 (76.5)
III	9 (39.1)	15 (32.6)		9 (52.9)	3 (17.6)
IV	2 (8.7)	5 (10.9)		0 (0)	1 (5.9)
Location			0.798			0.296
Proximal	13 (56.5)	23 (50.0)		12 (70.6)	8 (47.1)
Distal	10 (43.5)	23 (50.0)		5 (29.4)	9 (52.9)
Volume change portal phase (cm³)	37.903	26.945	0.168	34.034	20.152	0.140
Volume change delay phase (cm³)	37.887	27.069	0.213	33.557	23.766	0.324
Pre_AFP (≥8.78 ng/mL)	4 (17.4)	9 (19.6)	>0.99	4 (23.5)	4 (23.5)	>0.999
Pre_CEA (≥5 ng/mL)	8 (34.8)	19 (41.3)	0.794	9 (52.9)	5 (29.4)	0.296
Pre_CA-125 (≥35 u/mL)	9 (39.1)	12 (26.1)	0.405	5 (29.4)	3 (17.6)	0.686
Pre_CA-724 (≥8.2 u/mL)	13 (56.5)	18 (39.1)	0.266	9 (52.9)	8 (47.1)	>0.999
Pre_CA-199 (≥35 u/mL)	9 (39.1)	15 (32.6)	0.789	11 (64.7)	2 (11.8)	0.005
Post_AFP (≥8.78 ng/mL)	3 (13.0)	9 (19.6)	0.736	2 (11.8)	3 (17.6)	>0.999
Post_CEA (≥5 ng/mL)	4 (17.4)	11 (23.9)	0.757	3 (17.6)	3 (17.6)	>0.999
Post_CA-125 (≥35 u/mL)	0 (0.0)	2 (4.3)	0.800	1 (5.9)	0 (0.0)	>0.999
Post_CA-724 (≥8.2 u/mL)	5 (21.7)	19 (41.3)	0.180	3 (17.6)	7 (41.2)	0.259
Post_CA-199 (≥35 u/mL)	4 (17.4)	5 (10.9)	0.705	4 (23.5)	0 (0.0)	0.110

Values are given as n (%) or mean ± SD unless otherwise indicated.

AFP, alpha-fetoprotein; AJCC, American Joint Committee on Cancer; CA-125, carbohydrate antigen 125; CA-199, carbohydrate antigen 199; CA-724, carbohydrate antigen 724; CEA, carcinoembryonic antigen; SRCC, signet-ring cell carcinoma; TRG, tumor regression grade.

Feature selection and screening

After Spearman correlation analysis, 234 radiomics features presented as highly correlated. Therefore, the 234 radiomics features were considered robust and used for subsequent analysis. When performing dimensionality reduction by LASSO, three, five, and six radiomics features remained for the pre-NAC set, the post-NAC set, and the delta set, respectively (Supplemental Fig. S2). The tuned λ values using minimum criterion for the three feature sets were 0.1411, 0.1085, and 0.1136. Descriptive information for features in each radiomics set is shown in Fig. 2 and Supplemental Table S1.

Radiomic signature generation and modeling

When modeling for evaluation of TRG, the delta radiomics set performed the best in both cohorts (0.91 and 0.76 in the training cohort and validation cohort) (Fig. 3). No significance was found when comparing the AUCs of each model between the training cohort and validation cohort (P_pre−NAC = 0.515, P_post−NAC = 0.238, P_delta = 0.116). The delta model also achieved the highest accuracy among the models in the validation cohort (delta vs. pre-NAC vs. post-NAC = 70.6% vs. 52.9% vs. 52.9%). More information on the classification metrics of the models is presented in Table 2. All six radiomics features in the delta feature set came from delayed phase images, including two from mixed images and four from IU images. Multivariable logistic regression indicated only one feature (d_IU_wavelet.HLL_glszm_ZonePercentage, from IU images in the delayed phase) acted as an independent predictive factor in the delta model (OR = 6.26, 95% CI = 1.73–39.00; P = 0.019) and the remaining five radiomics features showed no statistical significance. Information for the three radiomics models is shown in Supplemental Table S1. Finally, the radiomics signature was chosen and calculated from the delta radiomics feature set. To further develop the clinical value of the delta model, we subjected patients’ clinical information (sex, age, cTNM stage, Borrmann type, tumor indices, etc.) and relevant immunohistochemical indices (Her-2, MLH1, MSH2, MSH6) to univariate analysis, which showed no significant difference in the above factors (Supplemental Table S2).

Table 2.

The performance of the three radiomics models for TRG evaluation.

	Accuracy (%)	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
Pre-NAC model	52.9	35.3	70.6	54.6	52.2
Post-NAC model	52.9	70.6	35.3	52.2	54.6
Delta model	70.6	82.4	58.8	66.7	76.9

NAC, neoadjuvant chemotherapy; NPV, negative predicted value; PPV, positive predicted value; TRG, tumor regression grade.

Performance of the radiomics signature for prognosis prediction

The Cox model constructed by the optimal radiomics signature showed significant capability to discriminate TRG patients for both DFS and OS for the 40 patients in the independent cohort. The radiomics signature was indicated as a risk factor for DFS and OS when performing univariate Cox analysis (DFS: HR = 4.59, 95% CI = 1.04–20.24, P = 0.044; OS: HR = 58.43, 95% CI = 1.64–2086, P = 0.026). The C-indices for the radiomics signature to predict DFS and OS were 0.644 and 0.727, respectively. The high-risk group with a higher radiomics signature indicated a significantly poorer prognosis than the low-risk group by Kaplan–Meier analysis with the log-rank test (DFS and OS, P = 0.027 and 0.014, respectively) (see Fig. 4).

Discussion

In the present study, we proposed a radiomics signature generated by delta radiomics features based on pre- and post-NAC DECT images, achieving good performance in distinguishing treatment response from the TRG system for patients with locally advanced GC. The radiomics signature was an independent risk factor for both DFS and OS when a univariable Cox model was performed. Furthermore, this model showed moderate performance with C-indices of 0.644 and 0.727 to predict DFS and OS, respectively. The radiomics signature could be an alternative tool for TRG evaluation in patients with locally advanced GC and provides potential prognostic information.

NAC before surgery is associated with improved survival for patients with GC (4,27). However, conventional TRG systems for evaluating treatment response require invasive resection species and do not benefit non-responders. Radiomics, a promising field that provides non-invasive biomarkers from many quantitative features extracted from medical images, has been widely used for GC (28,29). Previous studies have reported the value of radiomics in evaluating treatment response after NAC for patients with GC (9,30,31). However, intrinsic limitations in those studies may dilute the evidence of radiomics, including using pretreatment images only, lack of validation, segmentation on only one slice of tumor, and so on.

Several advantages may strengthen the insight of radiomics provided in our research. First, unlike the studies mentioned above applying pretreatment CT images alone, we adopted a more rigorous process to filter the optimal radiomics signature, by simultaneously using and comparing pre- and post-NAC CT images. Consequently, the delta radiomics signature performed the best. Delta radiomics reveals the difference in radiomics parameters before and after treatment, thus providing dynamic changes in radiomics during therapy. The role of delta radiomics has been investigated in a variety of diseases, such as pancreatic cancer, rectal cancer, high-grade osteosarcoma, and esophageal squamous cell carcinoma (32 –35). However, delta radiomics is rarely used for GC, except for one. A preliminary study with 23 patients with advanced GC indicated that delta radiomics features have the capability to discriminate response and non-response after NAC but lacked validation (36). Our results may indicate that delta radiomics is more reliable for capturing the therapeutic process for different features in delta radiomics selected by LASSO compared with pre- or post-NAC radiomics features.

The second piece of potential evidence is the comprehensive analysis of our study on different modalities. We used DECT images for radiomics analysis, and DECT is a novel imaging modality capturing material difference of elemental composition. Our previous study revealed that total IU from DECT has the capability to predict the response of locally advanced GC after NAC (37). In our study, we found that all six features in the delta feature set were from the delayed phase and four out of those were of IU origin. Some drugs in NAC can reduce the blood supply to the tumor, thus leading to ischemic necrosis (38), while iodine-containing contrast agents reach the tumor site through blood circulation; therefore, the amount of iodine-containing contrast agents at the tumor site can, to a certain extent, reflect the tumor's response to chemotherapy. Multivariable logistic regression analysis showed that the only feature (d_IU_wavelet.HLL_glszm_ZonePercentage) that presented significant difference was also observed in the iodine images. This feature discloses the coarseness of the texture by taking the ratio of the number of zones and number of voxels in the ROI, with higher values indicating a more delicate texture (39). Previous studies disclosed that this feature could differentiate invasive lung adenocarcinomas and minimally invasive adenocarcinomas and predict lymph node metastases of endometrial cancer (40,41). In this study, the response to NAC may indicate less heterogeneity of the tumor, reflecting a positive change in this feature that is consistent with what we have observed.

Last, we utilized the radiomics signature calculated from TRG to predict the prognosis for locally advanced GC in an independent cohort. Previous studies have investigated the prognostic value of radiomics in GC. Jiang et al. (11) applied a Lasso-Cox regression model to preoperatively predict the survival and chemotherapeutic benefits of 1591 patients with GC. They found that the radiomics signature was an independent risk factor for both DFS and OS and that the radiomics-clinical nomograms were of improved performance in the estimation of survival compared with clinicopathological nomograms. Another study also revealed that the radiomics signature from fluorine 18F positron emission tomography images was a powerful predictor of OS and DFS, and higher Rad scores of patients with GC were prone to benefit from chemotherapy (42). Unlike the above studies, the radiomics signature used to predict survival in our study was based on the evaluation of TRG, and we found that the radiomics signature was an independent risk factor for both DFS and OS. The integrated design was similar to a previous study (31), which showed that the radiomics signature generated from TRG evaluation was capable of predicting OS. However, several intrinsic drawbacks may have decreased the value of radiomics in that study, including a lack of comparison from different modalities and ROIs delineated only on the largest cross-sectional area. Furthermore, in our study, we showed the success of the radiomics signature on survival prediction in an independent cohort instead of the primary cohorts, which increased the generalization of our radiomics signature.

The present study has some limitations. This was a preliminary retrospective study, and the sample size was small from a single center. To avoid bias and improve generalization, prospective multicenter studies are needed. In addition, we performed survival prediction by radiomics indirectly from TRG evaluation. The direct value of radiomics for survival prediction needs further investigation. Furthermore, due to a lack of follow-up data in the primary 103 patients, we could not present the predictive value of the radiomics signature based on those patients.

In conclusion, the delta radiomics signature can serve as an alternative non-invasive tool for the evaluation of TRG for locally advanced GC, avoiding unnecessary cytotoxicity for non-responders. Radiomics provided potential prognostic information.

Supplemental Material

sj-docx-1-acr-10.1177_02841851221123971 - Supplemental material for Efficacy and prognostic value of delta radiomics on dual-energy computed tomography for gastric cancer with neoadjuvant chemotherapy: a preliminary study

Supplemental material, sj-docx-1-acr-10.1177_02841851221123971 for Efficacy and prognostic value of delta radiomics on dual-energy computed tomography for gastric cancer with neoadjuvant chemotherapy: a preliminary study by Lingyun Wang, Yong Chen, Jingwen Tan, Yingqian Ge, Zhihan Xu, Michael Wels and Zilai Pan in Acta Radiologica

Footnotes

Acknowledgements

We would like to thank Ping Li and Haoliang Sun from Shanghai Engineering Research Center for Broad Technologies and Applications for their help in understanding the technology.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Shanghai Science and Technology Commission Science and Technology Innovation Action Clinical Innovation Field, Medical Engineering Cross Research Foundation of Shanghai Jiaotong University, and National Natural Science Foundation of China (grant nos. 18411953000, YG2019ZDB09, 81771789, and 81771790).

ORCID iDs

Lingyun Wang

Yong Chen

Supplemental material

Supplemental material for this article is available online.

References

Bray

Ferlay

Soerjomataram

, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018;68:394-424.

Strong

Selby

, et al. Differences in gastric cancer survival between the U. S. and China. J Surg Oncol 2015;112:31-37.

Zong

Abe

Seto

, et al. The challenge of screening for early gastric cancer in China. Lancet 2016;388:2606.

Van Cutsem

Sagaert

Topal

, et al. Gastric cancer. Lancet 2016;388:2654-2664.

Langer

Becker

. Tumor regression grading of gastrointestinal cancers after neoadjuvant therapy. Virchows Arch 2018;472:175-186.

Becker

Langer

Reim

, et al. Significance of histopathological tumor regression after neoadjuvant chemotherapy in gastric adenocarcinomas: a summary of 480 cases. Ann Surg 2011;253:934-939.

Achilli

De Martini

Ceresoli

, et al. Tumor response evaluation after neoadjuvant chemotherapy in locally advanced gastric adenocarcinoma: a prospective, multicenter cohort study. J Gastrointest Oncol 2017;8:1018-1025.

Zhu

Sun

, et al. Comparison of five tumor regression grading systems for gastric adenocarcinoma after neoadjuvant chemotherapy: a retrospective study of 192 cases from National Cancer Center in China. BMC Gastroenterol 2017;17:41.

Zhang

Dai

, et al. Computed tomography-based radiomics for prediction of neoadjuvant chemotherapy outcomes in locally advanced gastric cancer: a pilot study. Chin J Cancer Res 2018;30:406-414.

10.

Lambin

Rios-Velazquez

Leijenaar

, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 2012;48:441-446.

11.

Jiang

Chen

Xie

, et al. Radiomics signature of computed tomography imaging for prediction of survival and chemotherapeutic benefits in gastric cancer. EBioMedicine 2018;36:171-182.

12.

Zhang

Tian

, et al. Prognostic value of computed tomography radiomics features in patients with gastric cancer following curative resection. Eur Radiol 2019;29:3079-3089.

13.

McCollough

Leng

, et al. Dual- and multi-energy CT: principles, technical approaches, and clinical applications. Radiology 2015;276:637-653.

14.

Zhou

, et al. Radiomics analysis of dual-energy CT-derived iodine maps for diagnosing metastatic cervical lymph nodes in patients with papillary thyroid cancer. Eur Radiol 2020;30:6251-6262.

15.

Dong

Fang

, et al. Dual-energy CT-based deep learning radiomics can improve lymph node metastasis risk prediction for gastric cancer. Eur Radiol 2020;30:2324-2333.

16.

Chen Y, Xi W, Yao W, et al. Dual-energy computed tomography-based radiomics to predict peritoneal metastasis in gastric cancer. Front Oncol 2021 May 14;11:659981.

17.

Fave

Zhang

Yang

, et al. Delta-radiomics features for the prediction of patient outcomes in non-small cell lung cancer. Sci Rep 2017;7:588.

18.

Wang

, et al. Multimodal radiomics model to predict treatment response to neoadjuvant chemotherapy for locally advanced rectal cancer. World J Gastroenterol 2020;26:2388-2402.

19.

Wang

Peng

Feng

, et al. Development and validation of a computed tomography-based radiomics signature to predict response to neoadjuvant chemotherapy for locally advanced gastric cancer. JAMA Netw Open 2021;4:e2121143.

20.

Becker

Mueller

Schulmacher

, et al. Histomorphology and grading of regression in gastric carcinoma treated with neoadjuvant chemotherapy. Cancer 2003;98:1521-1530.

21.

Wang L, Zhang Y, Chen Y, et al. The performance of a dual-energy CT derived radiomics model in differentiating serosal invasion for advanced gastric cancer patients after neoadjuvant chemotherapy: Iodine map combined with 120-kV equivalent mixed images. Front Oncol 2021 Jan 11;10:562945.

22.

Graser

Johnson

Chandarana

, et al. Dual energy CT: preliminary observations and potential clinical applications in the abdomen. Eur Radiol 2019;19:13-23.

23.

Albrecht

Trommer

Wichmann

, et al. Comprehensive comparison of virtual monoenergetic and linearly blended reconstruction techniques in third-generation dual-source dual-energy computed tomography angiography of the thorax and abdomen. Invest Radiol 2016;51:582-590.

24.

Sofue

Itoh

Takahashi

, et al. Quantification of cisplatin using a modified 3-material decomposition algorithm at third-generation dual-source dual-energy computed tomography: an experimental study. Invest Radiol 2018;53:673-680.

25.

Wels

Lades

Mühlberg

, et al. General purpose radiomics for multimodal clinical research. Proc SPIE 2019; 1095046.

26.

Zwanenburg

Vallières

Abdalah

, et al. The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping. Radiology 2020;295:328-338.

27.

Miao

Liu

Wang

, et al. Effect of neoadjuvant chemotherapy in patients with gastric cancer: a PRISMA-compliant systematic review and meta-analysis. BMC Cancer 2018;18:118.

28.

Sah

Owczarczyk

Siddique

, et al. Radiomics in esophageal and gastric cancer. Abdom Radiol (NY) 2019;44:2048-2058.

29.

Wang

Jin

. Radiomics approaches in gastric cancer: a frontier in clinical decision making. Chin Med J (Engl) 2019;132:1983-1989.

30.

Giganti

Marra

Ambrosi

, et al. Pretreatment MDCT-based texture analysis for therapy response prediction in gastric cancer: comparison with tumour regression grade at final histology. Eur J Radiol 2017;90:129-137.

31.

Sun

Chen

, et al. CT-based radiomics scores predict response to neoadjuvant chemotherapy and survival in patients with gastric cancer. BMC Cancer 2020;20:468.

32.

Jeon

Song

Chang

, et al. Delta-radiomics signature predicts treatment outcomes after preoperative chemoradiotherapy in rectal cancer. Ann Oncol 2019;4:iv24.

33.

Nasief

Hall

Zheng

, et al. Improving treatment response prediction for chemoradiation therapy of pancreatic cancer using a combination of delta-radiomics and the clinical biomarker CA19-9. Front Oncol 2019;9:1464.

34.

Hou

Dai

, et al. Biophysic constraint model using spatial registration of delta 18F-fluorodeoxyglucose positron emission tomography/computed tomography images for predicting radiation pneumonitis in esophageal squamous cell carcinoma patients receiving neoadjuvant chemoradiation. Onco Targets Ther 2019;12:6439-6451.

35.

Lin

Yang

Chen

, et al. A delta-radiomics model for preoperative evaluation of neoadjuvant chemotherapy response in high-grade osteosarcoma. Cancer Imaging 2020;20:7.

36.

Mazzei

Nardone

Di Giacomo

, et al. The role of delta radiomics in gastric cancer. Quant Imaging Med Surg 2018;8:719-721.

37.

Gao

Zhang

Yuan

, et al. Locally advanced gastric cancer: total iodine uptake to predict the response of primary lesion to neoadjuvant chemotherapy. J Cancer Res Clin Oncol 2018;144:2207-2218.

38.

Zhang

Tong

Zhang

, et al. Arsenic sulfide combined with JQ1, chemotherapy agents, or celecoxib inhibit gastric and colon cancer cell growth. Drug Des Devel Ther 2015;9:5851-5862.

39.

van Griethuysen

Fedorov

Parmar

, et al. Computational radiomics system to decode the radiographic phenotype. Cancer Res 2017;77:e104-e107.

40.

Weng

Zhou

Wang

, et al. A radiomics model for determining the invasiveness of solitary pulmonary nodules that manifest as part-solid nodules. Clin Radiol 2019;74:933-943.

41.

De Bernardi

Buda

Guerra

, et al. Radiomics of the primary tumour as a tool to improve 18F-FDG-PET sensitivity in detecting nodal metastases in endometrial cancer. EJNMMI Res 2018;8:86.

42.

Jiang

Yuan

, et al. Radiomic signature of 18F fluorodeoxyglucose PET/CT for prediction of gastric cancer survival and chemotherapeutic benefits. Theranostics 2018;8:5915-5928.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.69 MB