CT-based radiomics predicts CD38 expression: indirectly reflects clinical prognosis in epithelial ovarian cancer

doi:10.21203/rs.3.rs-3748054/v1

Background

Cluster of differentiation 38 (CD38) has been found to be highly expressed in various solid tumors, and its expression level may be associated with patient prognosis and survival. The study aimed to explore the prognostic value of CD38 expression for patients with epithelial ovarian cancer (EOC) and to construct two computed tomography (CT)-based radiomics models for CD38 expression prediction.

Methods

A total of 333 cases of EOC were enrolled from The Cancer Genome Atlas (TCGA) database for CD38 related bioinformatics and survival analysis. 56 intersection cases from TCGA and The Cancer Imaging Archive (TCIA) databases were selected for radiomics feature extraction and model construction. Logistic regression (LR) and support vector machine (SVM) models were constructed and internally validated using 5-fold cross-validation to assess the performance of the models for CD38 expression levels.

Results

High expression of CD38 was an independent protective factor (HR = 0.540) for overall survival (OS) in EOC patients. 5 radiomics features based on CT images were selected to build models for CD38 expression prediction. In the training set and internal validation set, for the receiver operating characteristic (ROC) Curve, the LR model reached area under the curve (AUC) of 0.739 and 0.732, while the SVM model achieved AUC values of 0.741 and 0.700, respectively. For the precision-recall (PR) Curve, the LR model and SVM model demonstrated AUC of 0.760 and 0.721. The calibration curves and decision curve analysis (DCA) provided evidence supporting the fitness and net benefit of the models.

Conclusions

High expression of CD38 can improve OS in EOC patients. CT-based radiomics models can be a new predictive tool for CD38 expression, offering possibilities for individualized survival assessment for patients with EOC.

CD38

Epithelial ovarian cancer

Radiomics

CT

Survival

Prognosis

Epithelial ovarian cancer (EOC) is the most common type of ovarian malignancy, accounting for over 95% of cases[1]. Annually, approximately 230,000 women are diagnosed with EOC, with more than 75% of patients being diagnosed at an advanced stage. Furthermore, this disease claims the lives of around 150,000 patients per year worldwide[2]. The standard treatment for EOC involves surgical intervention followed by platinum-based adjuvant chemotherapy. Recent advancements in immunotherapy and targeted therapy have emerged as promising treatment options[3]. However, despite these advances, the average 5-year survival rate for EOC patients remains below 50%, especially for those diagnosed at an advanced stage, where the 5-year survival rate is a mere 29%[1, 2]. Moreover, the prognosis and survival outcomes vary significantly among individuals [2, 4],attributed to the considerable heterogeneity observed both among patients and within tumors [1, 2]. The current diagnostic and prognostic indicators of EOC, including clinical and pathological characteristics[5], serum markers such as carbohydrate antigen 125 (CA125) and human epididymis protein 4 (HE4)[6], and traditional techniques like ultrasound, computed tomography (CT), and magnetic resonance imaging (MRI), are inadequate to meet the clinical demands of precision medicine[7]. Consequently, there is a pressing need to explore novel prognostic indicators that can facilitate early and effective assessment of patients and supporting clinical decision-making.

The cluster of differentiation 38 (CD38) protein encoded by the CD38 gene is a non-lineage-restricted type II transmembrane glycoprotein that synthesizes and hydrolyzes cyclic adenosine 5'-diphosphate ribose, serving as an intracellular calcium mobilization regulator. CD38 is expressed on the surface of various immune cells, promoting immune cell activation, proliferation, and adhesion[8], and also exerting certain regulatory effects on tumor development[9]. Studies have found that CD38 is also highly expressed in solid tumor tissues such as head and neck squamous cell carcinoma[10], lung cancer[11], cervical cancer[12], and prostate cancer[13], and has been confirmed to be associated with immune cell infiltration and tumor metabolism regulation. Furthermore, Zhu et al.[14] found that CD38 expression in EOC is higher than in normal tissues, and high CD38 expression has been reported to be associated with longer disease-free survival (DFS). Therefore, CD38 may have the potential as a biomarker for EOC prognosis and maybe as a therapeutic target as reported[15]. However, the current detection methods for CD38 protein or message ribonucleic acid (mRNA), such as peripheral blood cytokine detection, flow cytometry, ribonucleic acid sequencing (RNA-seq), and pathological immunohistochemistry, have some drawbacks, such as invasiveness, complexity of operation, and relatively high cost, making them unsuitable for widespread clinical use. Therefore, non-invasive and efficient methods that can reflect molecular expression levels are still being explored.

Currently, CT is one of the most commonly used imaging methods for diagnosis, staging, and follow-up of ovarian cancer (OC)[7]. CT images are easily accessible and provide valuable information on the morphology of the lesion, surrounding infiltration, and extent of metastasis. However, traditional imaging methods such as CT and MRI rely on visual identification and have limitations in providing comprehensive information[16]. Radiomics, on the other hand, has the potential to overcome these limitations by converting conventional medical image information into high-dimensional data that can be analyzed quantitatively. Radiomics provides a wide range of quantitative features based on intensity, shape, volume, and texture, which are closely related to molecular-level characteristics and the tumor microenvironment[16]. It has been successfully applied in various disease states for diagnosis, prognosis evaluation, pathological classification, staging, etc.[17–20]. In recent years, there has been growing interest in using radiomics to predict the expression levels of relevant disease molecular biomarkers and disease prognosis, showing potential promising clinical applications[21–24].

Therefore, based on the background, this study aims to investigate the correlation between CD38 expression levels and the survival status of patients with EOC, as well as explore preliminary potential mechanisms and immune microenvironment associations. We also attempt to construct two radiomics machine learning prediction models based on CT images and evaluate their feasibility for non-invasive prediction of CD38 expression levels in EOC tissues. The ultimate goal is to indirectly reflect the prognosis and survival of patients with EOC.

Data acquisition

Transcriptome sequencing data (including clinical and follow-up data) were obtained from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/) to investigate the bioinformatics features associated with CD38 and its prognostic value for patient survival. The inclusion criteria for the study were as follows: (1) patients with primary treatment ovarian serous cystadenocarcinoma (OSC), which is the main subtype of EOC[2]; (2) availability of RNA-seq data from the primary solid tumor. The exclusion criteria were as follows: (1) data with missing survival status or follow-up time were excluded; (2) data with missing clinical information, unknown pathological international federation of gynecology and obstetrics (FIGO) staging, unknown pathological grading were excluded; (3) data with a follow-up time of less than 30 days were excluded.

For survival analysis, the following seven variables were included as covariates: age (< 60 years vs ≥ 60 years), chemotherapy (yes vs no), FIGO stage (I/II vs III vs IV), lymphatic invasion (yes vs no vs unknown), tumor histological grade (G1/G2 vs G3/G4), tumor residual disease (No Macroscopic disease vs 1–10 mm vs ≥ 11mm vs Unknown), and vascular invasion (yes vs no vs unknown).

The plain and enhanced abdominal CT imaging data (including clinical and follow-up data) were downloaded from The Cancer Imaging Archive (TCIA) database (https://www.cancerimagingarchive.net/). The inclusion criteria involved abdominal CT imaging data that intersected with the selected TCGA database cases. Exclusion criteria were applied to remove imaging data with inadequate image quality, defined as images with significant artifacts or poor resolution. The final included TCGA-TCIA intersecting data was used for extracting radiomic features and constructing prediction models.

CD38 related bioinformatics and survival analysis

The cut-off value for CD38 mRNA expression levels was calculated using the "survminer" package in R software (version 4.2.1). The data were then divided into high-expression and low-expression groups, which were labeled as CD38^high and CD38^low groups, respectively. After processed through the Toil pipeline[25], the RNA data in TPM format for EOC from the TCGA dataset, along with the RNA data of corresponding normal tissue from the Genotype-tissue expression (GTEx) database, were downloaded and extracted from UCSC XENA (https://xenabrowser.net/datapages/). To compare the expression differences of CD38 between EOC tissues and normal tissues, the "limma" R package were utilized.

The gene expression matrix of EOC RNA-seq data was analyzed using the ImmuCellAI database (http://bioinfo.life.hust.edu.cn/ImmuCellAI/#!/) to assess immune cell infiltration status. The differential immune cell infiltration between CD38^high and CD38^low groups was compared. The "clusterProfiler" R package was utilized to perform gene set enrichment analysis (GSEA) on the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Hallmark gene sets. A bar plot visualizing the top 30 enriched pathways was generated. Additionally, the correlation between CD38 expression and apoptosis genes was also investigated.

The main outcome for clinical prognosis analysis was overall survival (OS). The R packages 'survival' was used to plot evaluate Kaplan-Meier survival curves, then the survival rates of CD38^high and CD38^low groups were evaluated. The difference of the survival rates between the two groups was assessed using the Log-Rank test. The 'survival' R package was also utilized to conduct univariate and multivariate Cox regression analyses, investigating the impact of various factors on the OS.

Images segmentation and radiomics feature extraction

The CT images were resampled to 1×1×1mm³ pixels to achieve isotropic resolution. The images were then standardized to minimize discrepancies in grayscale values resulting from image acquisition on different machines.

The lesion segmentation was performed using 3D Slicer software (version 4.10.2; https://www.slicer.org/). A radiologist (H Liu) with 12 years of expertise in gynecological imaging diagnosis, blinded to the CD38 expression levels and other clinical data, manually delineated the lesions layer by layer. Then, a random group of 10 cases was selected using the "random number table method" for secondary delineation by a senior radiologist (HJ Zhang) with 18 years of experience. The open-source "Pyradiomics" package (https://pyradiomics.readthedocs.io/en/latest) based on Python (version 3.7.4, https://www.python.org/downloads/) was employed to extract radiomic features. The extracted radiomic features were Z-score standardized using the R package "caret". The consistency of the radiomic features extracted from the volumes of interest (VOIs) delineated by the two radiologists was evaluated using the intraclass correlation coefficient (ICC) with the R package "irr". Features with an ICC ≥ 0.80 were selected for further selection.

Radiomics feature selection

The features with an ICC of 0.80 or higher were included. The maximum relevance minimum redundancy (mRMR) algorithm was employed to select ten features that have the highest relevance to the classification variable while minimizing redundancy among features. Next, we utilized the stepwise regression algorithm with the akaike information criterion (AIC) for further features selection. Finally, the selected features were included in the final set of radiomics features.

Construction and evaluation of prediction models

Two radiomics prediction models, logistic regression (LR) model and support vector machine (SVM) model, were constructed. The sigmoid function was used to map the linear combination of selected radiomics features to a value between 0 and 1, allowing for the establishment of a binary classification model for predicting CD38 expression. The 'stats' package in R software was utilized for LR model establishment.

The SVM algorithm, on the other hand, utilizes support vectors to find a hyperplane in high-dimensional space. This hyperplane maximizes the margin between the classification hyperplane and support vectors, thereby identifying the optimal decision boundary. The "caret" package in R was used to model the selected radiomics features using SVM.

The performance of the LR and SVM radiomics models was assessed using 5-fold cross-validation. The receiver operating characteristic (ROC) Curve was plotted and area under the curve (AUC) was calculated to evaluate the diagnostic performance. precision-recall (PR) curve was also generated to assess the discriminative and diagnostic ability, with corresponding AUC value calculated. Calibration curves were plotted and the Hosmer-Lemeshow goodness-of-fit test was used to evaluate calibration degree of predicted and observed results. The decision curve analysis (DCA) was performed to assess the clinical utility. The AUC values of the models before and after cross-validation were compared using the DeLong test. The analyses were conducted using R packages "pROC," "measures," "ResourceSelection," "rms," and "rmda."

The LR and SVM radiomics models output the probability of CD38 expression level, which is then designated as the radiomics score (Rad_score). The difference in Rad_scores between the CD38^high and CD38^low groups was evaluated.

Statistical analysis

Numerical data were reported as mean ± standard deviation, while categorical data were presented as frequencies and percentages. The baseline characteristics of categorical variables were compared using the chi-square test, whereas continuous variables were compared using the Wilcoxon test. The correlation between two factors was analyzed using Spearman's correlation coefficient. The differences in quantitative data between two groups were compared using the Wilcoxon test. Statistical analysis was conducted using R software (version 4.2.1).

Baseline characteristics

Based on the inclusion and exclusion criteria, a total of 333 patient data were included from TCGA database. The cut-off value for CD38 expression level was calculated as 0.757, and according to this criterion, the patients were divided into CD38^high group (145 cases) and CD38^low group (188 cases). The baseline characteristics of the included patients are shown in Table 1. Except for tumor residual disease (P=0.024), there were no significant differences in other clinical features between the two groups. After screening the TCIA database, a total of 56 TCGA-TCIA intersecting cases were finally included. The Inclusion-Exclusion Flowchart can be seen in Supplementary Figure 1.

Table 1 Clinical characteristics of the EOC patients with high and low CD38 expression

Variables	Total (n=333)	Low (n=188)	High (n=145)	P
Age, n (%)
<60	171(51)	100 (53)	71(49)	0.513
≥60	162(49)	88(47)	74(51)	0.513
FIGO stage, n (%)
I/II	18(5)	8(4)	10(7)	0.324
III	265(80)	148(79)	117(81)
IV	50(15)	32(17)	18(12)
Histologic grade, n (%)
G1/G2	41(12)	28(15)	13(9)	0.143
G3/G4	292(88)	160(85)	132(91)	0.143
Tumor residual disease, n (%)
No macroscopic disease	58(18)	34(18)	24(17)	0.024
1-10mm	161(48)	95(50)	66(46)
≥11mm	84(25)	50(27)	34(23)
Unknown	30(9)	9(5)	21(14)
Lymphatic invasion, n (%)
Yes	90(27)	45(24)	45(31)	0.127
No	39(12)	27(14)	12(8)
Unknown	204(61)	116(62)	88(61)
Venous invasion, n (%)
Yes	58(18)	30(16)	28(19)	0.510
No	31(9)	20(11)	11(8)
Unknown	244(73)	138(73)	106(73)
Chemotherapy, n (%)
Yes	312(94)	177(94)	135(93)	0.871
No	21(6)	11(6)	10(7)	0.871

Expression level, immune infiltrates and gene enrichment analysis of CD38

CD38 expression is significantly upregulated in EOC compared to normal tissues (p < 0.01), as shown in Figure 1-A.

Differences in the degree of immune cell infiltration were observed between the CD38^high and CD38^low groups. Specifically, the CD38^high group showed a significant increase in infiltration of some types of immune cells such as Natural Killer (NK) cells, Dendritic Cells (DC), and gamma-delta T (γδT) cells compared to the CD38^low group (p < 0.001), as shown in Figure 1-B.

The results of GSEA revealed significant enrichment of differentially expressed genes between the CD38^high and CD38^lowgroups in multiple signaling pathways, mainly including the mitogen-activated protein kinase (MAPK) signaling pathway, chemokine signaling pathway, apoptosis, and PI3K-AKT-mTOR signaling, as shown in Figure 1-C and D. CD38 expression is significantly positively correlated (p < 0.001) with CSF2RB, PIK3CG, and FASLG genes, as shown in Supplementary Figure 2.

CD38 related survival analysis

Figure 2-A illustrated the Kaplan-Meier survival curves for CD38^high and CD38^low groups. The median survival time for CD38^high group was 52.77 months, while for CD38^low group, it was 42.13 months, indicating a significant association between high CD38 expression and prolonged overall survival (P < 0.001).

The results of the univariate Cox regression analysis indicated that high expression of CD38 (HR=0.554, 95% CI: 0.415-0.740, P < 0.001) and chemotherapy (HR=0.470, 95% CI: 0.288-0.767, P = 0.003) were protective factors for OS. The results of the multivariate Cox regression analysis demonstrated that high expression of CD38 (HR =0.540, 95% CI: 0.400-0.730, P < 0.001) and chemotherapy (HR=0.377, 95% CI: 0.226-0.627, P < 0.001) were independent protective factors for OS. As shown in Figure 2-B. In addition, we performed COX subgroup analysis on the main variable CD38, which confirmed that there was no significant interaction between the covariates and the association of CD38 and OS. Details showed in Supplementary Figure 3.

Radiomics feature extraction and selection

The 56 TCGA-TCIA cases, were divided into CD38^high (29 cases) and CD38^low (27 cases) groups based on the cut-off value of 0.757. A total of 107 radiomics features were extracted from manually delineated tumor regions and standardized. The median value of the ICC was 0.963, with 101 features having an ICC ≥ 0.80, 5 features with 0.5 ≤ ICC ≤ 0.8, and 1 feature with ICC < 0.5. Features with an ICC ≥ 0.80 were further selected for subsequent feature selection. Finally, five radiomics features, including glcm_ldn, glrlm_RunLenthNonUniformity, gldm_DependenceVariance, glszm_LargeAreaHighGrayLevelEmphasis, gldm_SmallDependenceLowGrayLevelEmphasis, were selected by mRMR and stepwise regression.

Construction and evaluation of LR and SVM models

In the LR model, the overall importance of the selected radiomic features is illustrated in Supplementary Figure 4. The formula for predicting the probability of CD38 expression level, represented by the Rad_score, is as follows:

Rad_score = 0.220-0.547*(glcm_ldn)

-0.706*(glrlm_RunLenthNonUniformity)

-0.566*(gldm_DependenceVariance)

+1.787*(glszm_LargeAreaHighGrayLevelEmphasis)

-1.181*(gldm_SmallDependenceLowGrayLevelEmphasis)

The SVM model was also constructed using the aforementioned five radiomics features. The importance of each radiomics feature in the model is shown in Supplementary Figure 5.

The predictive performance of the LR and SVM models was evaluated. The LR model achieved an AUC value of 0.739 (95%CI:0.609-0.870) for the ROC curve, and after 5-fold cross-validation, the ROC-AUC of the LR model was 0.732 (95%CI:0.595-0.868) as shown in Figure 3-A and B. The SVM model had an AUC value of 0.741 (95%CI:0.608-0.874) for the ROC curve, and a 5-fold cross-validated ROC-AUC of 0.700(95%CI:0.555-0.835), as shown in Figure 4-A and B. Hosmer-Lemeshow goodness-of-fit test demonstrated comparatively good consistency between the predicted probabilities and the actual values for both the LR (P=0.838) and SVM (P=0.074) models, The corresponding calibration curves are shown in Figure 3-C and Figure 4-C. The AUC for the PR curve was 0.760 for the LR model (Figure 3-D) and 0.721 for the SVM model (Figure 4-D). The DCA curves showed if the risk threshold probability is between 30% to 80%, the two model both had superior net benefit (Figure 3-E and 4-E). The Delong test indicated no statistically significant difference in AUC values before cross-validation and after cross-validation in the two models (P=0.961 of LR model, P=0.703 of SVM model), indicating the stability of the model fitting.

The LR and SVM models both output the probability of predicting CD38 expression level, referred to as Rad_score. There was a significant difference in the distribution of Rad_score between the CD38^high and CD38^low groups (p<0.01). The CD38^high group exhibited higher Rad_score (Figure 3-F and 4-F).

Our research findings demonstrated that: (1) CD38 is highly expressed in EOC tissues; the expression level of CD38 is associated with immune cell infiltration, some tumor regulatory pathways, and microenvironment regulation; (2) High expression of CD38 is an independent protective factor of OS in EOC patients; (3) The two machine learning models based on CT radiomics features are able to predict the expression levels of CD38.

EOC exhibits high heterogeneity, that contributes to obvious differences in patients’ prognosis[26]. In recent years, research on molecular mechanisms related to tumor metabolism regulation and the tumor microenvironment has become a hot topic in accurately predicting the prognosis of EOC. Some relevant molecules may have potential value as prognostic biomarkers [27–30].

CD38 could serve as a biomarker for reflecting prognosis and survival. Some previous research[14, 31] and our study demonstrated the protective effect of high CD38 expression in improving the survival of EOC patients. Possible correlation with immune cell infiltration and signaling pathways have also been investigated, the CD38^high group exhibited increased infiltration of NK cells [32], γδT cells [33] with tumor-killing ability and immune regulation, as well as DC [34] that induce CD8 + T cell-mediated immune response [31, 35]. The differences observed between the CD38^high and CD38^low groups primarily involve signaling pathways regulating cell proliferation, growth, differentiation, apoptosis, and tumor metabolism. CD38 regulates and activates signaling pathways such as PI3_AKT_MTOR_SIGNALING has been confirmed [12].

In recent years, machine learning methods based on radiomics have been applied in various areas, including tumor diagnosis, differential diagnosis, staging, treatment response, and survival prediction, showcasing their promising diagnostic and predictive capabilities[36–40]. Several studies have endeavored to develop radiomics models utilizing CT, MRI, and other imaging modalities to predict biomarker expression levels in different tumor tissues. For instance, these models have successfully predicted the expression levels of phosphorylated β-arrestin in hepatocellular carcinoma [41], low-density lipoprotein receptor-related protein-1 in locally advanced rectal cancer [42], as well as C-C motif chemokine receptor 5 [22] and programmed cell death protein 1 [21] in ovarian cancer. Therefore, radiomics has shown sufficient value in non-invasive prediction of biomarker expression.

In our study, two radiomics models for CD38 expression prediction based on CT images of EOC patients were established. The AUCs of ROC both revealed fine predictive ability of LR model and SVM model. The PR curve also showed adequate discrimination ability of the two models. The LR model shows better predictive performance in comparison to the SVM model. Moreover, both models exhibited similar AUC values in both the training and validation sets, indicating consistent and stable performance. Additionally, calibration curves demonstrated a satisfactory consistency between the actual observed results and predicted value, DCA curves indicated a potential clinical net benefit for application.

To the best of our knowledge, this study may represent the first attempt to develop predictive radiomics models for assessing CD38 expression in patients with EOC based on CT imaging. CT imaging is widely utilized in clinical practice and the images are easily obtainable. The radiomics approach enables the extraction of quantitative features from conventional images, providing abundant objective information for further analysis. Therefore, the CT-based radiomics models offer a simple, non-invasive, and cost-effective method to predict the expression level of CD38. These models have the potential to indirect reflect the prognosis and survival outcomes of EOC patients. Furthermore, CD38, as a well-known therapeutic target for multiple myeloma[43], has been suggested as a potential target for solid tumors based on previous research[15]. In the future, by utilizing the radiomics model, EOC patients with high CD38 expression may potentially receive individualized targeted therapies at an early stage, leading to improved prognosis and survival.

The study has several limitations. Firstly, all the original images were obtained from the TCIA public database, which may introduce inherent variations between the images and potentially impact subsequent image analysis. Secondly, the sample size included in this retrospective study is relatively small, and the effectiveness of the model was evaluated using internal validation alone, lacking an independent external validation set. Furthermore, the manual delineation of VOIs is subjective, introducing potential biases that can affect subsequent analysis. Therefore, it is necessary to conduct multicenter studies with larger sample sizes. Additionally, incorporating more objective radiomics practice methods, such as automated or semi-automated VOI delineation, can enhance the accuracy and reliability of the analysis.

High expression of CD38 can improve OS in EOC patients. In this study, we developed CT-based radiomics machine learning models using LR and SVM algorithms to predict the expression level of CD38 in EOC tissue. The models, especially the LR model, exhibited fine predictive performance, providing potential for individualized prognosis assessment and clinical decision-making in patients with EOC.

CD38: Cluster of differentiation 38; EOC: Epithelial ovarian cancer; CT: Computed tomography; TCGA: The Cancer Genome Atlas; TCIA: The Cancer Imaging Archive; LR: Logistic regression; SVM: Support vector machine; OS: Overall survival; ROC: Receiver operating characteristic; AUC: Area under the curve; PR Curve: Precision-recall curve; DCA: Decision curve analysis; CA 125 Carbohydrate antigen 125; HE 4: Human epididymis protein 4; MRI: Magnetic resonance imaging; DFS: Disease-free survival; mRNA: Message ribonucleic acid; RNA-seq: Ribonucleic acid sequencing; OC: Ovarian cancer; OSC: Ovarian serous cystadenocarcinoma; FIGO: International federation of gynecology and obstetrics; GTEx database: Genotype-tissue expression database; GSEA: Gene set enrichment analysis; KEGG: The kyoto encyclopedia of genes and genomes; VOIs: The volumes of interest; ICC: Correlation coefficient; mRMR :Maximum relevance minimum redundancy; AIC: Akaike information criterion; Rad_score: Radiomics score; NK cell: Natural killer cell; DC: Dendritic cell; γδT cell: Gamma-delta T cell; MAPK signaling pathway: The mitogen-activated protein kinase signaling pathway.

Acknowledgements

We thank the study participants for permitting us to use their personal data.

Authors’ contributions

Literature search: Yuan Yao, Haijin Zhang, Hui Liu, Chendi Teng. Figures: Yuan Yao, Haijin Zhang, Wei Bian. Study design: Yuan Yao, Haijin Zhang, Xuan Che. Data collection: Haijin Zhang, Hui Liu, Chendi Teng, Wenting Zhang. Data analysis: Yuan Yao, Xuan Che, Haijin Zhang, Hui Liu. Writing of manuscript: Yuan Yao, Haijin Zhang. Review of manuscript: Xuan Che, Zhifeng Wang. The authors read and approved the final manuscript.

Funding

This study was funded by Jiaxing public welfare research project (No.2021AD30028).

Availability of data and materials

All data generated or analysed during this study are included in this published article and its supplementary information files.The datasets for this study were downloaded from TCIA (https:// www. cance rimag ingar chive. net/) and TCGA(https:// portal. gdc. cancer. gov/) database.

Ethics approval and consent to participate

Given that the TCGA and TCIA organizations have anonymized patients' private information and made their data publicly accessible, the requirement for ethical review and informed consent has been waived.

Consent for publication

Not required.

Competing interests

The authors declare that they have no competing interests.

Lheureux S, Braunstein M, Oza AM. Epithelial ovarian cancer: Evolution of management in the era of precision medicine. CA Cancer J Clin. 2019;69(4):280–304.
Lheureux S, Gourley C, Vergote I, Oza AM. Epithelial ovarian cancer. Lancet. 2019;393(10177):1240–53.
Kuroki L, Guntupalli SR. Treatment of epithelial ovarian cancer. BMJ. 2020;371:m3773.
Siegel RL, Miller KD, Wagle NS, Jemal A. Cancer statistics, 2023. CA Cancer J Clin. 2023;73(1):17–48.
Gaitskell K, Hermon C, Barnes I, Pirie K, Floud S, Green J, Beral V, Reeves GK. Million Women Study C: Ovarian cancer survival by stage, histotype, and pre-diagnostic lifestyle factors, in the prospective UK Million Women Study. Cancer Epidemiol. 2022;76:102074.
Rong Y, Li L. Early clearance of serum HE4 and CA125 in predicting platinum sensitivity and prognosis in epithelial ovarian cancer. J Ovarian Res. 2021;14(1):2.
Rizzo S, Del Grande M, Manganaro L, Papadia A, Del Grande F. Imaging before cytoreductive surgery in advanced ovarian cancer patients. Int J Gynecol Cancer. 2020;30(1):133–8.
Li W, Liang L, Liao Q, Li Y, Zhou Y. CD38: An important regulator of T cell function. Biomed Pharmacother. 2022;153:113395.
Gao L, Du X, Li J, Qin FX. Evolving roles of CD38 metabolism in solid tumour microenvironment. Br J Cancer. 2023;128(4):492–504.
Zhang MJ, Gao W, Liu S, Siu SP, Yin M, Ng JC, Chow VL, Chan JY, Wong TS. CD38 triggers inflammasome-mediated pyroptotic cell death in head and neck squamous cell carcinoma. Am J Cancer Res. 2020;10(9):2895–908.
Wu P, Zhao L, Chen Y, Xin Z, Lin M, Hao Z, Chen X, Chen D, Wu D, Chai Y. CD38 identifies pre-activated CD8 + T cells which can be reinvigorated by anti-PD-1 blockade in human lung cancer. Cancer Immunol Immunother. 2021;70(12):3603–16.
Liao S, Liang L, Yue C, He J, He Z, Jin X, Luo G, Zhou Y. CD38 is involved in cell energy metabolism via activating the PI3K/AKT/mTOR signaling pathway in cervical cancer cells. Int J Oncol. 2020;57(1):338–54.
Guo C, Crespo M, Gurel B, Dolling D, Rekowski J, Sharp A, Petremolo A, Sumanasuriya S, Rodrigues DN, Ferreira A, et al. CD38 in Advanced Prostate Cancers. Eur Urol. 2021;79(6):736–46.
Zhu Y, Zhang Z, Jiang Z, Liu Y, Zhou J. CD38 Predicts Favorable Prognosis by Enhancing Immune Infiltration and Antitumor Immunity in the Epithelial Ovarian Cancer Microenvironment. Front Genet. 2020;11:369.
Zucali PA, Lin CC, Carthon BC, Bauer TM, Tucci M, Italiano A, Iacovelli R, Su WC, Massard C, Saleh M et al. Targeting CD38 and PD-1 with isatuximab plus cemiplimab in patients with advanced solid malignancies: results from a phase I/II open-label, multicenter study. J Immunother Cancer 2022, 10(1).
Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563–77.
Bera K, Braman N, Gupta A, Velcheti V, Madabhushi A. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. Nat Rev Clin Oncol. 2022;19(2):132–46.
Feng Z, Li H, Liu Q, Duan J, Zhou W, Yu X, Chen Q, Liu Z, Wang W, Rong P. CT Radiomics to Predict Macrotrabecular-Massive Subtype and Immune Status in Hepatocellular Carcinoma. Radiology. 2023;307(1):e221291.
Li G, Li L, Li Y, Qian Z, Wu F, He Y, Jiang H, Li R, Wang D, Zhai Y, et al. An MRI radiomics approach to predict survival and tumour-infiltrating macrophages in gliomas. Brain. 2022;145(3):1151–61.
Qi Y, Zhao T, Han M. The application of radiomics in predicting gene mutations in cancer. Eur Radiol. 2022;32(6):4014–24.
Gao L, Jiang W, Yue Q, Ye R, Li Y, Hong J, Zhang M. Radiomic model to predict the expression of PD-1 and overall survival of patients with ovarian cancer. Int Immunopharmacol. 2022;113(Pt A):109335.
Wan S, Zhou T, Che R, Li Y, Peng J, Wu Y, Gu S, Cheng J, Hua X. CT-based machine learning radiomics predicts CCR5 expression level and survival in ovarian cancer. J Ovarian Res. 2023;16(1):1.
Meissner AK, Gutsche R, Galldiks N, Kocher M, Junger ST, Eich ML, Nogova L, Araceli T, Schmidt NO, Ruge MI, et al. Radiomics for the non-invasive prediction of PD-L1 expression in patients with brain metastases secondary to non-small cell lung cancer. J Neurooncol. 2023;163(3):597–605.
Ramtohul T, Djerroudi L, Lissavalid E, Nhy C, Redon L, Ikni L, Djelouah M, Journo G, Menet E, Cabel L, et al. Multiparametric MRI and Radiomics for the Prediction of HER2-Zero, -Low, and -Positive Breast Cancers. Radiology. 2023;308(2):e222646.
Vivian J, Rao AA, Nothaft FA, Ketchum C, Armstrong J, Novak A, Pfeil J, Narkizian J, Deran AD, Musselman-Brown A, et al. Toil enables reproducible, open source, big biomedical data analyses. Nat Biotechnol. 2017;35(4):314–6.
Reid F, Bhatla N, Oza AM, Blank SV, Cohen R, Adams T, Benites A, Gardiner D, Gregory S, Suzuki M, et al. The World Ovarian Cancer Coalition Every Woman Study: identifying challenges and opportunities to improve survival and quality of life. Int J Gynecol Cancer. 2021;31(2):238–44.
Banerjee S, Drapkin R, Richardson DL, Birrer M. Targeting NaPi2b in ovarian cancer. Cancer Treat Rev. 2023;112:102489.
Gahlawat AW, Witte T, Haarhuis L, Schott S. A novel circulating miRNA panel for non-invasive ovarian cancer diagnosis and prognosis. Br J Cancer. 2022;127(8):1550–6.
Ge L, Sun Y, Shi Y, Liu G, Teng F, Geng Z, Chen X, Xu H, Xu J, Jia X. Plasma circRNA microarray profiling identifies novel circRNA biomarkers for the diagnosis of ovarian cancer. J Ovarian Res. 2022;15(1):58.
Xi X, Cao T, Qian Y, Wang H, Ju S, Chen Y, Chen T, Yang J, Liang B, Hou S. CDC20 is a novel biomarker for improved clinical predictions in epithelial ovarian cancer. Am J Cancer Res. 2022;12(7):3303–17.
Li L, Chen D, Luo X, Wang Z, Yu H, Gao W, Zhong W. Identification of CD8(+) T Cell Related Biomarkers in Ovarian Cancer. Front Genet. 2022;13:860161.
Pugh-Toole M, Nicolela AP, Nersesian S, Leung BM, Boudreau JE. Natural Killer Cells: the Missing Link in Effective Treatment for High-Grade Serous Ovarian Carcinoma. Curr Treat Options Oncol. 2022;23(2):210–26.
Saura-Esteller J, de Jong M, King LA, Ensing E, Winograd B, de Gruijl TD, Parren P, van der Vliet HJ. Gamma Delta T-Cell Based Cancer Immunotherapy: Past-Present-Future. Front Immunol. 2022;13:915837.
Rob L, Cibula D, Knapp P, Mallmann P, Klat J, Minar L, Bartos P, Chovanec J, Valha P, Pluta M et al. Safety and efficacy of dendritic cell-based immunotherapy DCVAC/OvCa added to first-line chemotherapy (carboplatin plus paclitaxel) for epithelial ovarian cancer: a phase 2, open-label, multicenter, randomized trial. J Immunother Cancer 2022, 10(1).
Chang H, Zhu Y, Zheng J, Chen L, Lin J, Yao J. Construction of a Macrophage Infiltration Regulatory Network and Related Prognostic Model of High-Grade Serous Ovarian Cancer. J Oncol 2021, 2021:1331031.
Avanzo M, Stancanello J, Pirrone G, Sartor G. Radiomics and deep learning in lung cancer. Strahlenther Onkol. 2020;196(10):879–87.
Yu Y, He Z, Ouyang J, Tan Y, Chen Y, Gu Y, Mao L, Ren W, Wang J, Lin L, et al. Magnetic resonance imaging radiomics predicts preoperative axillary lymph node metastasis to support surgical decisions and is associated with tumor microenvironment in invasive breast cancer: A machine learning, multicenter study. EBioMedicine. 2021;69:103460.
Autorino R, Gui B, Panza G, Boldrini L, Cusumano D, Russo L, Nardangeli A, Persiani S, Campitelli M, Ferrandina G, et al. Radiomics-based prediction of two-year clinical outcome in locally advanced cervical cancer patients undergoing neoadjuvant chemoradiotherapy. Radiol Med. 2022;127(5):498–506.
Fasmer KE, Hodneland E, Dybvik JA, Wagner-Larsen K, Trovik J, Salvesen O, Krakstad C, Haldorsen IHS. Whole-Volume Tumor MRI Radiomics for Prognostic Modeling in Endometrial Cancer. J Magn Reson Imaging. 2021;53(3):928–37.
Bhandari A, Ibrahim M, Sharma C, Liong R, Gustafson S, Prior M. CT-based radiomics for differentiating renal tumours: a systematic review. Abdom Radiol (NY). 2021;46(5):2052–63.
Che F, Xu Q, Li Q, Huang ZX, Yang CW, Wang LY, Wei Y, Shi YJ, Song B. Radiomics signature: A potential biomarker for beta-arrestin1 phosphorylation prediction in hepatocellular carcinoma. World J Gastroenterol. 2022;28(14):1479–93.
Li Z, Huang H, Wang C, Zhao Z, Ma W, Wang D, Mao H, Liu F, Yang Y, Pan W, et al. DCE-MRI radiomics models predicting the expression of radioresistant-related factors of LRP-1 and survivin in locally advanced rectal cancer. Front Oncol. 2022;12:881341.
Gozzetti A, Ciofini S, Simoncelli M, Santoni A, Pacelli P, Raspadori D, Bocchia M. Anti CD38 monoclonal antibodies for multiple myeloma treatment. Hum Vaccin Immunother. 2022;18(5):2052658.

No competing interests reported.

Supplementarymaterials.docx

CT-based radiomics predicts CD38 expression: indirectly reflects clinical prognosis in epithelial ovarian cancer

Status:

Journal Publication

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Background

Methods

Data acquisition

CD38 related bioinformatics and survival analysis

Images segmentation and radiomics feature extraction

Radiomics feature selection

Construction and evaluation of prediction models

Statistical analysis

Results

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1