Radiomics modeling to predict the tumor immune-microenvironment of mucinous adenocarcinoma of gastric-type cervical cancer

Radiomics modeling to predict the tumor immune-microenvironment of mucinous adenocarcinoma of gastric-type cervical cancer | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Radiomics modeling to predict the tumor immune-microenvironment of mucinous adenocarcinoma of gastric-type cervical cancer Risa Matsuda, Kohei Oguma, Hiroshi Nishio, Yutaka Shiraishi, Masafumi Sawada, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7895813/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Gastric-type adenocarcinoma (GAS) of the cervix is usually diagnosed at advanced stages and has a poor prognosis due to resistance to standard therapies. Although immune checkpoint inhibitors have improved outcomes in cervical cancer, prognosis and treatment response are strongly influenced by the tumor immune microenvironment (TME). Histopathological assessment of GAS is challenging because it often arises in the upper cervix. This study aimed to predict the TME in GAS using MRI-based radiomics. We enrolled 16 patients with GAS treated at our institution. Tumor-infiltrating lymphocytes (TILs) were evaluated by immunohistochemistry, quantifying them with the Immunoscore. Fourteen patients with usual endocervical adenocarcinoma (UEA) served as controls. A total of 1,309 radiomic features were extracted from the primary tumor and peritumoral region on pre-treatment MRI images. After feature selection, clustering, and regression models were developed to predict the TME in GAS. GAS exhibited significantly lower T-cell infiltration than UEA, particularly in early-stage tumors. The clustering model achieved 87.5% accuracy in Immunoscore classification, and the regression model showed a strong correlation with observed TIL densities (r = 0.93, P < .001). These findings suggest that MRI-based radiomics may serve as a noninvasive biomarker for predicting the TME in GAS. Health sciences/Biomarkers Biological sciences/Cancer Biological sciences/Computational biology and bioinformatics Health sciences/Oncology cervical cancer gastric-type adenocarcinoma radiomics magnetic resonance imaging imaging analysis immune microenvironment Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Gastric-type adenocarcinoma (GAS) of the cervix is a distinct histological subtype of human papillomavirus (HPV)-independent cervical adenocarcinoma characterized by differentiation into gastropyloric glands[ 1 – 3 ]. Although GAS is rare, accounting for approximately 2–5% of cervical adenocarcinomas globally, it represents a higher proportion, about 20%, in Japan[ 4 ]. GAS is associated with poor prognosis due to its tendency to present at an advanced stage and its aggressive clinical behavior, including deep local invasion and frequent distant metastases[ 5 ]. GAS has also been reported to show resistance to conventional chemotherapy (e.g., platinum- and taxane-based regimens) and radiation therapy compared with usual-type endocervical adenocarcinoma (UEA)[ 5 ]. The tumor immune microenvironment (TME) has been recognized as a pivotal therapeutic target in advanced cervical cancer. The introduction of immune checkpoint inhibitors (ICIs) has significantly improved clinical outcomes, as demonstrated by the KEYNOTE-826 trial[ 6 ]. However, the immune profile of the TME in GAS remains poorly understood, largely because of the rarity of this disease. While immunotherapy, particularly ICIs, has become an established treatment option across multiple cancers, identifying reliable biomarkers to predict treatment efficacy remains essential. Previous studies have shown that the density of tumor-infiltrating lymphocytes (TILs) within the TME correlates with prognosis and ICI response. In cervical cancer, higher densities of CD8 + T cells within the tumor are associated with improved ICI responsiveness[ 7 ]. The Immunoscore, validated by Pagès et al.[ 8 ], quantifies CD3 + and CD8 + T-cell density within the tumor and has proven to be a strong prognostic marker, even surpassing TNM staging system in colorectal cancer. Furthermore, the Immunoscore has shown predictive value for ICI response across multiple malignancies[ 9 ]. Importantly, validated data demonstrate that the Immunoscore applies to biopsy specimens as well as resected tumors, supporting its potential relevance in GAS, which is often diagnosed at advanced, unresectable stages. Although no studies have yet evaluated the Immunoscore in GAS, it may provide valuable prognostic information and guide treatment strategies. GAS frequently arises in the upper cervix and is often diagnosed at advanced stages[ 10 ], making it difficult to obtain adequate tumor tissue for comprehensive analysis. Histopathological evaluation based solely on biopsy specimens has limitations, as it may not fully capture tumor heterogeneity. Imaging-based evaluations may therefore contribute to a more complete assessment. In this study, we aimed to develop a predictive model for the TME in GAS using imaging analysis. Radiomics, a technique that extracts and analyzes a large number of quantitative features from medical imaging modalities such as computed tomography or magnetic resonance imaging (MRI), has gained increasing use in oncology. It has been reported to predict malignancy, histological subtypes, driver gene mutations, treatment response, and prognosis across multiple cancer types[ 11 – 15 ]. In endometrial cancer, for example, MRI-based radiomic features have been shown to predict myometrial invasion and lymphovascular space invasion (LVSI)[ 16 , 17 ]. In cervical cancer, MRI radiomics has been applied to predict response to concurrent chemoradiotherapy[ 18 ]. The objective of our study was to explore the potential of MRI-based radiomics in predicting the immune status of GAS. We characterized the TME of GAS in comparison with UEA and further performed radiomics analysis focused on GAS, a setting in which tissue-based evaluation is often limited and imaging-based approaches may have particular clinical value. Methods Patient selection Patients with a pathological diagnosis of GAS who were treated at our institution between December 2018 and February 2023 were enrolled in this study. The inclusion criteria were: (1) clinical stage greater than stage IB1 (FIGO 2018), (2) presence of an identifiable tumor on pre-treatment MRI, and (3) no prior chemotherapy, immunotherapy, or radiotherapy before diagnosis. Patients who underwent pre-treatment MRI at other hospitals were also included. A total of 16 patients met the eligibility criteria and were included in the analysis. Descriptive statistics were calculated for all study variables. For comparison of T-cell infiltration, 14 patients with UEA were selected as controls. To minimize potential confounding from differences in tumor progression, UEA patients were manually matched to the GAS cohort by tumor size, depth of invasion, and clinical stage distribution. Collected clinical variables included age, year of diagnosis, clinical stage (FIGO 2018), tumor size, depth of stromal invasion, LVSI, paracervical invasion, vaginal wall invasion, lymph node metastasis, distant metastasis, and treatment received. Survival outcomes included follow-up duration after surgery, recurrence status, and vital status at the last follow-up. Tumor morphology on MRI was classified according to the proportion of cystic components: tumors with ≥ 50% cystic area were categorized as cystic-dominant, and those with < 50% as solid-dominant. Immunohistochemistry, TIL quantification, and calculation of Immunoscore Archived tissue samples from 16 patients with GAS and 14 with UEA, collected during routine clinical care, were used for analysis. Formalin-fixed, paraffin-embedded tissue blocks were serially sectioned at 4 µm. Sections were deparaffinized, rehydrated, and subjected to antigen retrieval by heating in a water bath at 95°C for 20 minutes. After blocking endogenous peroxidase activity, the sections were stained with antibodies against CD3 (790–4341, Roche), CD4 (790–4423, Roche), CD8 (M7103, Dako), and FoxP3 (ab20034, Abcam). They were incubated overnight at 4°C with monoclonal primary antibody, followed by amplification using a labeled polymer/HRP system (pv-6001, Zhongshan Golden Bridge Biotechnologies) at 37°C for one hour. Visualization was achieved with 3,3'-diaminobenzidine (DAB), and counterstaining was performed with hematoxylin. Digital images of stained slides were scanned at 40× magnification with a resolution of 0.2 µm/pixel. StrataQuest version 7 software (TissueGnostics) was used to quantify CD3 and CD8 expression. Tumor regions were manually segmented, and CD3 + and CD8 + cells were quantified based on DAB-positive nuclei. Thresholding was applied to ensure accurate detection for each core, and results were reported as cells/mm 2 . The Immunoscore methodology established by Pagès et al.[ 8 ] was adapted for this study. CD3 + and CD8 + cell densities in the tumor region were compared across all patients and converted to percentiles. Given that GAS frequently lack a discrete mass and often exhibit infiltrative growth with indistinct margins, and that some cases were represented only by biopsy specimens, differentiation between tumor center and invasive margin was not consistently feasible. Therefore, T-cell infiltration was assessed across the entire tumor region. The mean of the CD3 + and CD8 + percentiles was used to classify patients into two categories: low (< 25%), intermediate-to-high (25–100%). Progression-free survival (PFS) was then compared between Immunoscore groups among patients with GAS. Radiomic feature extraction Figure 1 illustrates the workflow of radiomics analysis. Three-dimensional radiomic features were extracted from the primary tumor (PT) and the surrounding peritumoral region (PR) using axial and sagittal T2-weighted images (T2WIs) and axial diffusion-weighted imaging (DWI) obtained at diagnosis. The PT was delineated by a radiation oncologist (Y.S.) with over 20 years of experience. To characterize the PR, 10 volumes of interest (VOIs) were generated by isotropically expanding the tumor margin by 2, 4, 6, 8, and 10 mm, with each expansion defining a peritumoral volume (i.e., the margin excluding the tumor itself). All MRI images underwent preprocessing, including intensity standardization, resizing, and intensity binning (see the supplementary material). For each combination of MRI sequence (three types) and VOI (11 types), 1,302 radiomic features were extracted using PyRadiomics version 3.0.28. These included 18 first-order features, 75 texture features from the original images, and 1,209 filtered features from 13 image transformations (eight wavelet decompositions combining high- and low-pass filters along the x, y, and z axes, and five Laplacian of Gaussian filters with σ values of 1, 2, 3, 4, and 5 mm). To evaluate interobserver reproducibility, a second radiation oncologist (M.S.) with over 10 years of experience, blinded to the initial delineations, independently delineated the PT. Radiomic features were extracted using the same procedure. Features with an intraclass correlation coefficient (ICC) > 0.75 were considered reproducible and included in subsequent analyses. Radiomic feature selection associated with tumor immune microenvironment The Immunoscore, validated by Pagès et al.,[ 8 ] is a percentile-based metric that stratifies immune cell densities into categorical scores. However, this transformation introduces non-linearity, making it less suitable as a target variable in linear regression models due to potential instability and reduced reproducibility. To address this limitation, we introduced a continuous alternative—the TIL density index—defined as the mean density of CD3 + and CD8 + cells (cells/mm²). This index was used as a surrogate for immune status, and radiomic models were developed to predict its value. Feature preselection was performed as follows: (1) the log-transformed TIL density index was used as the target variable to improve normality and ensure linearity for regression analysis. (2) All radiomic features were standardized using Z-score normalization. (3) Features showing a Pearson correlation of ( \(\:\left|{r}_{\text{y}}\right|>0.5\) ) with the target variable were selected as candidate predictors. (4) To reduce redundancy, Pearson correlation coefficients were calculated between all candidate features. If two features had a very strong inter-feature correlation ( \(\:\left|{r}_{\text{f}}\right|>0.9\) ), the feature with the lower absolute correlation with the target variable (lower \(\:\left|{r}_{\text{y}}\right|\) ) was excluded. Radiomic feature map visualization To examine the spatial distribution of radiomic features and improve interpretability, feature maps were generated using the feature with the highest reproducibility and strongest correlation with the log-transformed TIL density index. These maps were visually assessed to identify tumor regions contributing most to prediction, and findings were compared with histopathological evaluations of cell distribution in the tumor center, margin, and adjacent non-tumor areas. Developing a radiomics-based prediction model of tumor immune microenvironment For clustering, an unsupervised model was constructed by applying Principal component analysis (PCA) to reduce the preselected radiomic features to two principal components, followed by K-means clustering to stratify patients according to immune status (low vs. intermediate–high). For regression, a multivariable linear model was developed to predict the log-transformed TIL density index, using features further selected from the preselected set by the least absolute shrinkage and selection operator (LASSO). The LASSO regularization parameter ( \(\:\lambda\:\) ) was optimized via grid search (Supplementary Fig. S1 ). Statistical analysis Statistical analyses of patient characteristics, outcomes, and TIL infiltration were performed using SPSS (version 29.0; SPSS Inc., Chicago, IL, USA). Statistical significance was set at P < .05. Continuous variables are expressed as medians, with group comparisons performed using the Mann–Whitney U test. Categorical variables were analyzed using the χ 2 test or Fisher’s exact test. PFS was analyzed with the Kaplan–Meier method and log-rank test. Graphs were generated using GraphPad Prism (version 10.6.0). Statistical analyses for feature selection and model development were performed using Python (version 3.9.7) and R (version 4.4.2). Interobserver reproducibility was assessed by ICC with the irr package (version 0.84.1) in R. PCA and K-means clustering were conducted with the decomposition and cluster modules of scikit-learn (version 1.6.1) in Python. The clustering model was evaluated by accuracy, precision, recall, specificity, and F1 score. LASSO regression was performed using the linear_model module of scikit-learn. Regression model performance was assessed by the Pearson correlation coefficient, coefficient of determination (R²), and mean absolute error (MAE). Multicollinearity was considered acceptable when the variance inflation factor (VIF) was < 10, calculated with the stats module of SciPy (version 1.10.1) in Python. Ethical declarations This retrospective study was approved by the Institutional Review Board (IRB) of Keio University Hospital (No. 20030107 and 20110275), with informed consent waived due to its retrospective design. Use of tissue samples was approved by the IRB, and an opt-out consent process was applied in accordance with institutional and national ethical regulations. All methods were carried out in accordance with the relevant guidelines and regulations. Results Patient characteristics Of the 16 patients with GAS included in this study, the median age at diagnosis was 52 years (range, 33–70 years). Six patients (37.5%) were diagnosed at Stage IB, while the remaining 10 (62.5%) were at Stage II or higher (Table 1 ). The median tumor size was 31 mm (range, 9–60 mm), and the median depth of stromal invasion was 11 mm (range, 1–25 mm). The median follow-up duration was 22 months (range, 8–59 months). Of the 16 patients, 10 (62.5%) underwent surgical treatment, 4 (25.0%) received concurrent chemoradiotherapy, and 2 (12.5%) received chemotherapy alone. Among the 14 patients with UEA, the median age at diagnosis was 38 years (range, 29–52 years). The overall clinicopathological characteristics were comparable between the two groups, with no significant differences observed in tumor size or depth of invasion. GAS cases exhibited variable MRI morphologies, with 7 of 16 (43.8%) showing cystic-dominant features (≥ 50% cystic component) and 9 of 16 (56.2%) showing solid-dominant features. In contrast, all UEA cases (14/14) demonstrated solid morphology. Table 1 Patient characteristics: GAS n = 16 UEA n = 14 p value Age (years), median (range) 52 (33–70) 38 (29–52) 0.001 a Stage (FIGO2018) (n, %) IB 6 (37.5) 9 (64.3) 0.272 b II-IV 10 (62.5) 5 (35.7) LVSI (n, %) 5/10 (50.0) 10/12 (83.3) 0.172 c Tumor size (mm), median (range) 31 (9–60) 25 (17–55) 0.3179 Depth of invasion (mm), median (range) 11 (1–25) 10 (5–27) 1.000 a Lymph node metastasis (n, %) 5 (31.3) 4 (28.6) 1.000 c Parametrial invasion (n, %) 8 (50.0) 2 (14.3) 0.058 c Vaginal invasion (n, %) 4 (25.0) 3 (21.4) 1.000 c Lesion morphology (n, %) Cystic-dominant 7 (43.8) 0 (0) 0.007 c Solid-dominant 9 (56.2) 14 (100) Recurrence (n, %) 9 (56.3) 2 (14.3) 0.026 c Death (n, %) 6 (37.5) 1 (7.1) 0.086 c Observation period (months), median (range) 22 (8–59) 34 (10–50) 0.400 a GAS; gastric-type adenocarcinoma, UEA; usual endocervical adenocarcinoma, LVSI; lymphvascular space invasion a; Mann-Whitney U test, b; chi-square test, c; Fischer’s exact test Tumor immune microenvironment in GAS Immunohistochemical analysis was performed to evaluate TILs, with results shown in Fig. 2 . Representative cases of early-stage and advanced-stage GAS and UEA are presented in Fig. 2 a. A clear difference in TIL density between GAS and UEA was observed. GAS tumors exhibited lower infiltration of CD3 + and CD8 + T cells, and FoxP3-positive cells were also fewer in GAS (Fig. 2 a, b). GAS cases showed a trend toward a higher frequency of Immunoscore-low tumors (37.5% vs. 7.1%, P = 0.086) (Fig. 2 c). Figure 2 d compares immune status by clinical stage in GAS and UEA, assessed using the TIL density index. GAS tumors exhibited a distinct pattern, with markedly reduced T-cell infiltration, particularly in early-stage disease (TIL density index: 243 cells/mm² in early-stage vs. 875 cells/mm² in advanced stage; P < 0.001). By contrast, UEA cases demonstrated T-cell infiltration from early stages, with no significant variation across clinical stages. Among GAS patients, those with a high Immunoscore had significantly poorer prognosis compared with those with a low Immunoscore (Fig. 2 e). Immunoscore-associated radiomic features A total of 12 radiomic features were selected through the preselection process as candidate predictors of immune status (Table 2 ). The number of features selected at each step, stratified by image type and VOI, is summarized in Supplementary Fig. S2. RunEntropy from the gray-level run-length matrix (GLRLM), extracted from the PT volume with a 10-mm peritumoral margin on the original axial DWI sequence ( AxialDWI_PT + PR10mm_Original_GLRLM_RunEntropy ), showed high reproducibility (ICC = 0.773) and the strongest correlation with the log-transformed TIL density index ( \(\:{r}_{\text{y}}\) = 0.69, P = 0.003). RunEntropy quantifies the uncertainty or randomness in the distribution of run lengths and gray levels within the VOI; higher values indicate greater textural heterogeneity. This feature also showed consistent reproducibility and predictive performance when extracted from the PT volume with an additional peritumoral margin of 4–10 mm (Supplementary Fig. S3). However, when evaluated using the peritumoral volume alone, the feature did not meet reproducibility thresholds and was excluded. Table 2 Radiomic features associated with tumor immune microenvironment Modality Volume of interest Image filter Feature category Feature name r y p value Axial DWI PT + PR10mm Original GLRLM RunEntropy 0.69 0.003 Axial DWI PT + PR6mm LoG (σ = 1mm) Firstorder Mean -0.65 0.007 Axial T2WI PT LoG (σ = 2mm) GLCM InverseDifferenceNormalized 0.59 0.015 Axial T2WI PT + PR10mm LoG (σ = 4mm) GLSZM SmallAreaLowGrayLevelEmphasis -0.58 0.019 Axial DWI PT + PR8mm Original GLCM InformationalMeasureOfCorrelation2 0.57 0.021 Sagittal T2WI PR4mm wavelet-HLL Firstorder Skewness 0.56 0.025 Axial DWI PT + PR2mm LoG (σ = 1mm) GLCM MaximalCorrelationCoefficient 0.55 0.028 Sagittal T2WI PR2mm LoG (σ = 1mm) GLDM SmallDependenceLowGrayLevelEmphasis 0.53 0.036 Axial T2WI PT wavelet-LHL Firstorder Median 0.52 0.041 Sagittal T2WI PT + PR4mm wavelet-HLL GLRLM LongRunHighGrayLevelEmphasis -0.52 0.041 Axial DWI PT + PR2mm wavelet-LHL GLSZM HighGrayLevelZoneEmphasis 0.51 0.043 Axial DWI PT Original NGTDM Strength 0.51 0.044 r y = Pearson’s correlation coefficient with the log-transformed average density of CD3-positive and CD8-positive cells; DWI = Diffusion-Weighted Imaging; T2WI = T2-Weighted Imaging; PT = Primary Tumor; PRxmm = peritumoral region with an isotropic margin of x mm; LoG-σxmm = Laplacian of Gaussian filter with σ = x mm; wavelet-XYZ = wavelet-decomposed image in which each letter (X, Y, Z) denotes low-pass (L) or high-pass (H) filtering applied along the corresponding image axis; GLCM = Gray-Level Co-occurrence Matrix; GLRLM = Gray-Level Run-Length Matrix; GLSZM = Gray-Level Size-Zone Matrix; GLDM = Gray-Level Dependence Matrix; NGTDM = Neighborhood Gray-Tone Difference Matrix. Radiomic feature mapping We generated feature maps of DWIax_GTV + 10mm_original_GLRLM_RunEntropy , one of the most reproducible radiomic features and the one most strongly correlated with TIL density index. Representative MRI images, corresponding radiomic feature maps, and histologic cell density are shown in Fig. 3 . Cases 1 and 2 represent tumors with cystic morphology. The cystic regions in Case 1 demonstrated high signal intensity on DWI, whereas those in Case 2 appeared DWI-low. These differences were reflected in the RunEntropy values: Case 1 showed high and heterogeneous RunEntropy within the cystic regions, whereas Case 2 exhibited uniformly low values. Cases 3 and 4 represent tumors with solid morphology. In these cases, regions with high RunEntropy values were typically located at the tumor margins rather than the center (Fig. 3 a). Histopathological analysis confirmed that total cell density was highest in the tumor center, decreased at the tumor margin, and was lowest in non-tumor tissue, with marked heterogeneity at the tumor margins. The distribution of CD3 + T-cell density also aligned with the regions of high total cell density, as shown in the heatmap (Fig. 3 b–e). Radiomics-based prediction of tumor immune microenvironment The performance of the clustering approach is summarized in Fig. 4 a–c. The clustering model achieved an accuracy of 0.875, precision of 0.750, recall of 1.000, specificity of 0.800, and an F1 score of 0.857, with two false-negative cases misclassified. PCA of the 12 preselected features revealed that the first principal component (PC1) was positively correlated with the log-transformed TIL density index ( \(\:{r}_{\text{y}}\) = 0.86, P < 0.001). The contribution of DWIax_GTV + 10mm_original_GLRLM_RunEntropy accounted for 11.4% of the variance in PC1, the largest among the 12 features, making it the most influential radiomic feature for PC1 (see Supplementary Fig. S4). The performance of the regression model is shown in Fig. 4 d. Multivariable linear regression demonstrated strong predictive performance, with a Pearson correlation coefficient of 0.93 (P < 0.001) between predicted and observed values of the log-transformed TIL density index, an R² of 0.86, and a MAE of 0.12. Eight radiomic features were selected from the 12 preselected features by LASSO regression and are summarized in Supplementary Table S1 . Although all selected features showed statistically significant associations in univariate analysis, none remained significant in the multivariable model, likely due to the limited sample size. All VIFs were < 10, indicating minimal multicollinearity. Discussion Comparison with UEA revealed that GAS exhibits a distinctly different TME, characterized by sparse lymphocytic infiltration and an immunologically "cold" phenotype. Interestingly, in GAS, more advanced tumors tend to display increased immune activity compared with early-stage cases, suggesting potential susceptibility to immunotherapy. The limited immunogenicity of early GAS, likely due to highly differentiated morphology with minimal atypia and weak antigenicity, may hinder the initiation of immune responses and contribute to tumor invasion and progression. In other cancer types, immunologically "hot" tumors are often associated with better prognosis[ 8 , 19 – 21 ], which contrasts with the trend observed in GAS. This discrepancy may reflect the tendency for low TIL infiltration in early-stage disease and high TIL infiltration in advanced stages. In this study, we analyzed TIL infiltration patterns in GAS and established an MRI-based radiomics model to predict immune status. These results indicate that radiomics has potential as a noninvasive approach for assessing the TME in GAS. To validate the reliability of the radiomics prediction model and enhance interpretability, we examined the correspondence between extracted radiomic features and underlying histopathological architecture. RunEntropy , the most reproducible and TIL density-correlated radiomic feature in our study, is a texture feature derived from the gray-level run length matrix. It quantifies variability in the lengths of consecutive runs of pixels with the same intensity. It measures the randomness or complexity of these run lengths, with higher values indicating greater heterogeneity within the tissue. Histopathologically, increased RunEntropy has been associated with diverse tumor microarchitectures, including variations in cellular density, glandular structures, and cystic or necrotic areas. GAS can exhibit at least two distinct morphological patterns: (1) tumors predominantly composed of macrocystic structures, and (2) tumors mainly consisting of solid components. Our findings highlight that radiomic features derived from both cystic and solid components provide important information for predicting the Immunoscore. In cystic-dominant tumors, variations in DWI signal within the cystic components (Cases 1 and 2 in Fig. 3 a) likely reflect differences in tumor-secreted mucin profiles. Low-viscosity fluid with low protein content typically appears hypointense on DWI, whereas hemorrhagic or protein-rich mucin tends to appear hyperintense. Supporting this, prior reports have shown that during progression from lobular endocervical glandular hyperplasia to GAS, the positivity rates of MUC6, a marker of gastric-type mucin, and αGlcNAc, a carbohydrate modification of gastric mucin, decrease[ 22 ]. This suggests that alterations in mucin composition accompany tumor progression and may influence the immune microenvironment. In such tumors, radiomic features reflecting signal intensity or internal heterogeneity on DWI may correspond to cystic spaces containing serous or mucinous fluid of varying viscosity and protein content. These fluid characteristics could underlie the observed signal differences and may contribute to an immunosuppressive TME. In contrast, solid-dominant tumors, as shown in Cases 3 and 4 in Fig. 3 a, were characterized by higher RunEntropy values at the tumor margin, which corresponded histologically to regions with heterogeneous cell density and irregular lymphocyte infiltration. CD3 immunostaining confirmed that tumor centers often exhibited uniform, dense infiltration, whereas the periphery demonstrated more variable immune cell distribution. These findings suggest that radiomic features extracted from tumor margins—where signal heterogeneity and irregular interfaces are frequently observed—may reflect stromal reaction, lymphocytic infiltration, or tumor–stroma interactions, all critical components of the local immune contexture. Strengths of this study include the use of routinely acquired MRI scans without reliance on contrast-enhanced sequences or PET-MRI, allowing broad applicability across clinical settings and imaging protocols. GAS frequently exhibits poorly defined margins and adjacent organ invasion, which can compromise segmentation reproducibility and reduce the accuracy of radiomics-based models. To address this, we performed manual segmentation by two experienced radiologists and extracted only highly reproducible features, enhancing the robustness of our analysis. Another strength lies in the noninvasive, three-dimensional nature of MRI-based radiomics, which allows comprehensive assessment of the entire tumor—a particular advantage in GAS, where tissue sampling is often limited. With continued accumulation of clinical and imaging data, this approach holds promise for improving preoperative assessment, predicting therapeutic responses, and guiding personalized treatment strategies in GAS. Since TME is heterogeneous within tumors, accurate pathological assessment of the Immunoscore ideally requires evaluation of the entire resected specimen. However, prior studies of Immunoscore in colorectal cancer have included a substantial number of endoscopic biopsy samples, permitting evaluation on limited tissue[ 8 ]. Similarly, in GAS, many advanced cases are not amenable to resection. Therefore, consistent with prior reports, we included Immunoscore assessment based on biopsy samples. This may have introduced some uncertainty in Immunoscore evaluation in advanced compared with early-stage cases. Another limitation is the potential risk of overfitting in radiomics modeling due to the limited sample size. To mitigate this risk, we employed strategies, including feature preselection, reproducibility assessment using ICC, correlation filtering, and LASSO regularization. While these approaches reduce model complexity and improve robustness, external validation with larger, multi-institutional cohorts will be essential to confirm generalizability. Although the Immunoscore shows promise as a predictive biomarker for immunotherapy responsiveness, immune landscapes vary across organs and cancer types. An absolute cutoff for Immunoscore has not been established for cervical cancer. Further investigations are needed to determine appropriate thresholds, validate prognostic and predictive relevance, and assess clinical applicability in decision-making for patients with cervical cancer. In this study, we characterized the T-cell infiltration profile of GAS and successfully developed an MRI-based radiomics model capable of predicting immune status. Our findings suggest that radiomics may serve as a promising noninvasive tool for evaluating the TME in GAS. Declarations Additional Information Competing interests The authors declare no competing interests. Funding This study was supported by JSPS KAKENHI Grant Number JP25K02782 (Grant-in-Aid for Scientific Research (B)) and JP25K10545 (Grant-in-Aid for Scientific Research (C)). Author Contribution Risa Matsuda and Hiroshi Nishio contributed to study concept and design; Risa Matsuda and Kohei Oguma drafted the manuscript; Hiroshi Nishio obtained funding and supervised the study; Risa Matsuda and Hiroshi Nishio collected and analyzed clinical data; Kohei Oguma performed radiomics analysis; Miyuki Saito conducted immunohistochemistry and automated cell quantification; Maho Kurihara, Masafumi Sawada, and Yutaka Shiraishi contributed to radiodiagnosis and segmentation; Yutaka Shiraishi and Hiroshi Nishio provided senior supervision; Masaki Sugawara, Tomoya Matsui, and Takashi Iwata contributed to data interpretation and critical manuscript review; Masahiro Jinzaki, Atsuya Takeda, and Wataru Yamagami provided expert guidance as departmental chairs; all authors reviewed and approved the final version of the manuscript. Risa Matsuda and Kohei Oguma contributed equally to this study. Data Availability The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. References Kusanagi, Y. et al. Absence of high-risk human papillomavirus (HPV) detection in endocervical adenocarcinoma with gastric morphology and phenotype. Am. J. Pathol. 177 , 2169–2175. 10.2353/ajpath.2010.100323 (2010). Mikami, Y. Gastric-type mucinous carcinoma of the cervix and its precursors - historical overview. Histopathology 76 , 102–111. 10.1111/his.13993 (2020). Stolnicu, S. et al. Diagnostic Algorithmic Proposal Based on Comprehensive Immunohistochemical Evaluation of 297 Invasive Endocervical Adenocarcinomas. Am. J. Surg. Pathol. 42 , 989–1000. 10.1097/PAS.0000000000001090 (2018). Nishio, S. et al. Analysis of gastric-type mucinous carcinoma of the uterine cervix - An aggressive tumor with a poor prognosis: A multi-institutional study. Gynecol. Oncol. 153 , 13–19. 10.1016/j.ygyno.2019.01.022 (2019). Kojima, A. et al. Gastric morphology and immunophenotype predict poor outcome in mucinous adenocarcinoma of the uterine cervix. Am. J. Surg. Pathol. 31 , 664–672. 10.1097/01.pas.0000213434.91868.b0 (2007). Tewari, K. S. et al. Pembrolizumab or Placebo Plus Chemotherapy With or Without Bevacizumab for Persistent, Recurrent, or Metastatic Cervical Cancer: Subgroup Analyses From the KEYNOTE-826 Randomized Clinical Trial. JAMA Oncol. 10 , 185–192. 10.1001/jamaoncol.2023.5410 (2024). Wang, Y. et al. Multiparametric immune profiling of advanced cervical cancer to predict response to programmed death-1 inhibitor combination therapy: an exploratory study of the CLAP trial. Clin. Transl Oncol. 25 , 256–268. 10.1007/s12094-022-02945-1 (2023). Pages, F. et al. International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study. Lancet 391 , 2128–2139. 10.1016/S0140-6736(18)30789-X (2018). Eljilany, I. et al. The T Cell Immunoscore as a Reference for Biomarker Development Utilizing Real-World Data from Patients with Advanced Malignancies Treated with Immune Checkpoint Inhibitors. Cancers (Basel) . 15 10.3390/cancers15204913 (2023). Kido, A. et al. Magnetic resonance appearance of gastric-type adenocarcinoma of the uterine cervix in comparison with that of usual-type endocervical adenocarcinoma: a pitfall of newly described unusual subtype of endocervical adenocarcinoma. Int. J. Gynecol. Cancer . 24 , 1474–1479. 10.1097/IGC.0000000000000229 (2014). Gevaert, O. et al. Predictive radiogenomics modeling of EGFR mutation status in lung cancer. Sci. Rep. 7 , 41674. 10.1038/srep41674 (2017). Pesapane, F. et al. How Radiomics Can Improve Breast Cancer Diagnosis and Treatment. J. Clin. Med. 12 10.3390/jcm12041372 (2023). Son, J., Lee, S. E., Kim, E. K. & Kim, S. Prediction of breast cancer molecular subtypes using radiomics signatures of synthetic mammography from digital breast tomosynthesis. Sci. Rep. 10 , 21566. 10.1038/s41598-020-78681-9 (2020). Tagliafico, A. S. et al. Overview of radiomics in breast cancer diagnosis and prognostication. Breast 49 , 74–80. 10.1016/j.breast.2019.10.018 (2020). Wu, Y. J., Wu, F. Z., Yang, S. C., Tang, E. K. & Liang, C. H. Radiomics in Early Lung Cancer Diagnosis: From Diagnosis to Clinical Decision Support and Education. Diagnostics (Basel) . 12. 10.3390/diagnostics12051064 (2022). Han, Y. et al. Predicting myometrial invasion in endometrial cancer based on whole-uterine magnetic resonance radiomics. J. Cancer Res. Ther. 16 , 1648–1655. 10.4103/jcrt.JCRT_1393_20 (2020). Ma, W. et al. Predictive value of models based on MRI radiomics and clinical indicators for lymphovascular space invasion in endometrial cancer. BMC Cancer . 25 , 796. 10.1186/s12885-025-14217-6 (2025). Cai, C. et al. Longitudinal dynamic MRI radiomic models for early prediction of prognosis in locally advanced cervical cancer treated with concurrent chemoradiotherapy. Radiat. Oncol. 19 , 181. 10.1186/s13014-024-02574-8 (2024). Yun, S. et al. Immunoscore is a strong predictor of survival in the prognosis of stage II/III gastric cancer patients following 5-FU-based adjuvant chemotherapy. Cancer Immunol. Immunother . 70 , 431–441. 10.1007/s00262-020-02694-6 (2021). Zeng, L. et al. Clinical Significance of a CD3/CD8-Based Immunoscore in Neuroblastoma Patients Using Digital Pathology. Front. Immunol. 13 , 878457. 10.3389/fimmu.2022.878457 (2022). Schoumacher, C. et al. CD3-CD8 immune score associated with a clinical score stratifies PDAC prognosis regardless of adjuvant or neoadjuvant chemotherapy. Oncoimmunology 13 , 2294563. 10.1080/2162402X.2023.2294563 (2024). Yamanoi, K., Ishii, K., Tsukamoto, M., Asaka, S. & Nakayama, J. Gastric gland mucin-specific O-glycan expression decreases as tumor cells progress from lobular endocervical gland hyperplasia to cervical mucinous carcinoma, gastric type. Virchows Arch. 473 , 305–311. 10.1007/s00428-018-2381-6 (2018). Additional Declarations No competing interests reported. Supplementary Files 20251003SupplementaryMaterialimmunoscoreGASSciRep.pdf Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 03 Feb, 2026 Reviewers agreed at journal 29 Jan, 2026 Reviewers invited by journal 14 Nov, 2025 Editor invited by journal 28 Oct, 2025 Editor assigned by journal 21 Oct, 2025 Submission checks completed at journal 21 Oct, 2025 First submitted to journal 18 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7895813","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":549676061,"identity":"a21fff9a-5a9d-4618-abf7-10d9bd30f8ca","order_by":0,"name":"Risa Matsuda","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Risa","middleName":"","lastName":"Matsuda","suffix":""},{"id":549676062,"identity":"763465cb-b3b1-4a29-9dc6-f635ff905e00","order_by":1,"name":"Kohei Oguma","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Kohei","middleName":"","lastName":"Oguma","suffix":""},{"id":549676063,"identity":"ae1690ab-9db0-4830-8739-79a228a7ee88","order_by":2,"name":"Hiroshi Nishio","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAu0lEQVRIiWNgGAWjYDACZgY2hgQgLc/exgBiMBOvxbDnGLFaGIBawOBGGpHuMmdnf/bgQQWDPePMZ4kgBrs5IS2WzTzmBglnGJjZpdMOgxmWDQS0GBzmYZNIbGNgY5yd3gZiMBscIKiF/RlIJQ/DzePtP4jUwmAG0iLBcIPtGAORWnjMJBLOSBgY9qQlgxhE+OX88WeSPyps7OXZjxl+BDKSCYYYFEjAGckGRGpBADvStYyCUTAKRsFwBwBYFzaB16BdEQAAAABJRU5ErkJggg==","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Hiroshi","middleName":"","lastName":"Nishio","suffix":""},{"id":549676064,"identity":"a53063eb-47ec-407b-98d5-be03c0ce8b93","order_by":3,"name":"Yutaka Shiraishi","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yutaka","middleName":"","lastName":"Shiraishi","suffix":""},{"id":549676065,"identity":"e4b144b3-3efc-4d04-b309-e6898fd9e914","order_by":4,"name":"Masafumi Sawada","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Masafumi","middleName":"","lastName":"Sawada","suffix":""},{"id":549676066,"identity":"88bb80a4-6b04-4ec4-949d-9bac49bd1073","order_by":5,"name":"Maho Kurihara","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Maho","middleName":"","lastName":"Kurihara","suffix":""},{"id":549676067,"identity":"150a264a-9848-4152-9e7f-311c866a0e5a","order_by":6,"name":"Masaki Sugawara","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Masaki","middleName":"","lastName":"Sugawara","suffix":""},{"id":549676068,"identity":"6e70f28d-1c21-432d-8b83-1a4123617205","order_by":7,"name":"Miyuki Saito","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Miyuki","middleName":"","lastName":"Saito","suffix":""},{"id":549676069,"identity":"8a6fedda-b55e-4024-9578-9509185ed226","order_by":8,"name":"Tomoya Matsui","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Tomoya","middleName":"","lastName":"Matsui","suffix":""},{"id":549676070,"identity":"70372533-ff6a-41f1-988b-c3bb6678978c","order_by":9,"name":"Iwata Takashi","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Iwata","middleName":"","lastName":"Takashi","suffix":""},{"id":549676071,"identity":"25f82d35-72b6-4ea5-b1b6-5210d526fc30","order_by":10,"name":"Masahiro Jinzaki","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Masahiro","middleName":"","lastName":"Jinzaki","suffix":""},{"id":549676072,"identity":"b660a648-bf8a-4cb8-8aa9-cf9f6cbc9993","order_by":11,"name":"Wataru Yamagami","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Wataru","middleName":"","lastName":"Yamagami","suffix":""},{"id":549676073,"identity":"4db6d563-e24b-452d-99d7-d2d814341296","order_by":12,"name":"Atsuya Takeda","email":"","orcid":"","institution":"Keio University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Atsuya","middleName":"","lastName":"Takeda","suffix":""}],"badges":[],"createdAt":"2025-10-19 00:53:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7895813/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7895813/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96916362,"identity":"6f41e7a3-69e7-4504-9456-402de8dadf26","added_by":"auto","created_at":"2025-11-27 14:08:29","extension":"tif","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3703526,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure1immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/8e281e262bbdd272fa29b3f4.tif"},{"id":96792226,"identity":"3cc152e5-31cc-4001-a59b-5f462dd78320","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":115197,"visible":true,"origin":"","legend":"","description":"","filename":"20251013immunoscoreGASSciRep.docx","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/7cea50bda9a05a60436beb32.docx"},{"id":96792227,"identity":"50656f4b-f239-4218-83bf-d3d5b50b48f8","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8009132,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure2immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/9c4e063d5e00c91b56b1345b.tif"},{"id":96792236,"identity":"ddc2b1b3-c4d2-4443-9c7b-375083d6fd4d","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12147028,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure3immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/a6e7d0b69b43851d95c8d07d.tif"},{"id":96916620,"identity":"b19bfa67-b3bf-40cf-bfb4-87d5089dd2e2","added_by":"auto","created_at":"2025-11-27 14:08:47","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3226130,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure4immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/a979a18b6ff954ab01b04161.tif"},{"id":96792229,"identity":"09fd6470-713a-4890-8e1f-20e4ec5e94a9","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"json","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13028,"visible":true,"origin":"","legend":"","description":"","filename":"29ecad64e9d24918a7b95a4c157327e2.json","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/81c7469180cb3378e299724f.json"},{"id":96792242,"identity":"07714146-93e6-4178-b7fd-6d570b4d32a7","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":482484,"visible":true,"origin":"","legend":"","description":"","filename":"20251003SupplementaryMaterialimmunoscoreGASSciRep.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/d5661b374297603c0c9ee23d.pdf"},{"id":96915942,"identity":"6e2a087e-d0c2-43e2-bc9e-a5c871f5866f","added_by":"auto","created_at":"2025-11-27 14:07:48","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96443,"visible":true,"origin":"","legend":"","description":"","filename":"29ecad64e9d24918a7b95a4c157327e21enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/04a56461cf01b80e1ff4524a.xml"},{"id":96916143,"identity":"e7e92e2c-fae3-4b28-b2f3-ea04fb776e40","added_by":"auto","created_at":"2025-11-27 14:08:05","extension":"tif","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3703526,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure1immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/b13d89086660299b8a0274f3.tif"},{"id":96792239,"identity":"2e866b25-5022-4daa-a434-d2d557c8568c","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"tif","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8009132,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure2immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/83a981b5c3e59ca38119f86c.tif"},{"id":96792244,"identity":"b8914df4-7322-47d7-8d40-f452858edf8c","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"tif","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12147028,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure3immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/1d9598daec1e02e1ef355107.tif"},{"id":96792245,"identity":"92d44cb4-b864-4af3-aaf2-d8c858b364e3","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"tif","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3226130,"visible":true,"origin":"","legend":"","description":"","filename":"20251003Figure4immunoscoreGASSciRep.tif","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/ae921fc99fcbf46d23bad484.tif"},{"id":96792230,"identity":"63e7555c-0bd4-4c3e-ba05-6fc6f1cae39b","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":588039,"visible":true,"origin":"","legend":"","description":"","filename":"Online20251003Figure1immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/f69d4e412e78c0b77871131f.png"},{"id":96792246,"identity":"a1603c95-4944-4b9c-9882-b4846af7f43f","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1306355,"visible":true,"origin":"","legend":"","description":"","filename":"Online20251003Figure2immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/093b2375e0f79ee774bee636.png"},{"id":96916282,"identity":"15e5b8aa-2361-4418-aab3-d7ef431570be","added_by":"auto","created_at":"2025-11-27 14:08:24","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1196470,"visible":true,"origin":"","legend":"","description":"","filename":"Online20251003Figure3immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/db5216c3194d23b7d5a66451.png"},{"id":96792240,"identity":"9d0e80e6-0822-4154-9dd9-7254f1782a6b","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":406357,"visible":true,"origin":"","legend":"","description":"","filename":"Online20251003Figure4immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/3eccf88cefc5ea4514e18f91.png"},{"id":96792247,"identity":"403982d4-ea5c-4e2e-ac90-1527e196f633","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"xml","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96371,"visible":true,"origin":"","legend":"","description":"","filename":"29ecad64e9d24918a7b95a4c157327e21structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/95f84a20da97e00cb3f6581a.xml"},{"id":96792249,"identity":"c8ee3665-630b-47f5-ad93-6fe605633070","added_by":"auto","created_at":"2025-11-26 07:01:38","extension":"html","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":106705,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/03c33938a2b9cd833a2059e6.html"},{"id":96792232,"identity":"21ab086b-419f-4992-a6e3-909e2f4fb963","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":25572559,"visible":true,"origin":"","legend":"\u003cp\u003eMulti-parametric MRI-based radiomics workflow for predicting immune status in gastric-type adenocarcinoma\u003c/p\u003e","description":"","filename":"20251003Figure1immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/33a14f5092d5d9bafb48f5f2.png"},{"id":96916075,"identity":"bef224fa-ac42-4207-b08a-f162329d3e2c","added_by":"auto","created_at":"2025-11-27 14:07:59","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":34021055,"visible":true,"origin":"","legend":"\u003cp\u003eTumor immune microenvironment in gastric-type endocervical adenocarcinoma (GAS) compared to usual endocervical adenocarcinoma (UEA). \u003cstrong\u003ea\u003c/strong\u003e: Representative immunohistochemical staining of early-stage and advanced-stage GAS and UEA. \u003cstrong\u003eb\u003c/strong\u003e: Quantitative comparison of T-cell infiltration between GAS and UEA. \u003cstrong\u003ec\u003c/strong\u003e: Immunoscore-based classification of GAS and UEA cases. \u003cstrong\u003ed\u003c/strong\u003e: Comparison of TIL density index across clinical stages. \u003cstrong\u003ee\u003c/strong\u003e: Kaplan–Meier progression-free survival curves of GAS patients stratified by Immunoscore (low vs. intermediate–high). H\u0026amp;E; Haematoxylin and eosin. * p\u0026lt;0.05, ** p\u0026lt;0.01, NS; no significant difference.\u003c/p\u003e","description":"","filename":"20251003Figure2immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/039eacfc3a0b9f9f5d5c3270.png"},{"id":96916670,"identity":"396face1-d782-428e-a33c-4f398565bc4d","added_by":"auto","created_at":"2025-11-27 14:08:49","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":50911833,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative cases demonstrating the relationship between radiomic feature maps and histopathology. \u003cstrong\u003ea\u003c/strong\u003e: Cases 1 and 2 show tumors with a cystic morphology. The cystic region in Case 1 exhibited high and heterogeneous RunEntropy values, whereas Case 2 showed uniformly low values. Cases 3 and 4 display solid tumors, with high RunEntropy signals localized predominantly at the tumor margins. \u003cstrong\u003eb–e\u003c/strong\u003e: Histopathological comparison of cell density in tumor center, tumor margin, and non-tumor regions. Both total cells and CD3+ cells display high, uniform density in the tumor center, heterogeneous density at the margins, and low density in the surrounding non-tumor tissue. \u003cstrong\u003eb\u003c/strong\u003e: Hematoxylin and eosin staining. \u003cstrong\u003ec\u003c/strong\u003e: Heatmap of total cell density. \u003cstrong\u003ed\u003c/strong\u003e: Annotated map of tumor-infiltrating T cells. \u003cstrong\u003ee\u003c/strong\u003e: Heatmap of CD3+ T-cell density.\u003c/p\u003e","description":"","filename":"20251003Figure3immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/cfb00c0e7fdcb12502c5710a.png"},{"id":96792234,"identity":"921c639e-c49f-4143-b8b6-b0c6f4451e61","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":19127837,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance of clustering and regression models for immune status prediction using radiomic features. \u003cstrong\u003ea\u003c/strong\u003e: Principal component analysis (PCA) plot of radiomics features. Dots indicate patients colored by immune status (blue: Low, red: Intermediate–high). The shaded background shows clustering results (blue: Low, red: Intermediate–high), and asterisks denote misclassified cases between clustering and true status. \u003cstrong\u003eb\u003c/strong\u003e: PCA plot with radiomic feature maps of gray-level run-length matrix (GLRLM) RunEntropy extracted from the primary tumor volume with a 10 mm margin on the original axial diffusion-weighted imaging (DWI) image. \u003cstrong\u003ec\u003c/strong\u003e: Confusion matrix of clustering results versus true immune status. \u003cstrong\u003ed\u003c/strong\u003e: Scatter plot of predicted versus actual values of ln(TIL density index) showing the performance of a multivariable regression model using radiomics features [Pearson’s r = 0.93 (p \u0026lt; 0.001), R² = 0.86, and mean absolute error = 0.12]. The dashed line indicates the ideal relationship.\u003c/p\u003e","description":"","filename":"20251003Figure4immunoscoreGASSciRep.png","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/19d1756ae722768d6c7e4db8.png"},{"id":96923148,"identity":"0d093aa4-4e08-4cc1-ad3c-267d365a9bf7","added_by":"auto","created_at":"2025-11-27 14:20:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":106417225,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/d6b9291f-a261-4139-bcc3-d08a22c65954.pdf"},{"id":96792233,"identity":"c704bbc8-4356-4ce5-9877-8b3e267bc228","added_by":"auto","created_at":"2025-11-26 07:01:37","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":482484,"visible":true,"origin":"","legend":"","description":"","filename":"20251003SupplementaryMaterialimmunoscoreGASSciRep.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7895813/v1/97a1e78e586b38311f942b52.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Radiomics modeling to predict the tumor immune-microenvironment of mucinous adenocarcinoma of gastric-type cervical cancer","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGastric-type adenocarcinoma (GAS) of the cervix is a distinct histological subtype of human papillomavirus (HPV)-independent cervical adenocarcinoma characterized by differentiation into gastropyloric glands[\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Although GAS is rare, accounting for approximately 2\u0026ndash;5% of cervical adenocarcinomas globally, it represents a higher proportion, about 20%, in Japan[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. GAS is associated with poor prognosis due to its tendency to present at an advanced stage and its aggressive clinical behavior, including deep local invasion and frequent distant metastases[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eGAS has also been reported to show resistance to conventional chemotherapy (e.g., platinum- and taxane-based regimens) and radiation therapy compared with usual-type endocervical adenocarcinoma (UEA)[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The tumor immune microenvironment (TME) has been recognized as a pivotal therapeutic target in advanced cervical cancer. The introduction of immune checkpoint inhibitors (ICIs) has significantly improved clinical outcomes, as demonstrated by the KEYNOTE-826 trial[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. However, the immune profile of the TME in GAS remains poorly understood, largely because of the rarity of this disease.\u003c/p\u003e\u003cp\u003eWhile immunotherapy, particularly ICIs, has become an established treatment option across multiple cancers, identifying reliable biomarkers to predict treatment efficacy remains essential. Previous studies have shown that the density of tumor-infiltrating lymphocytes (TILs) within the TME correlates with prognosis and ICI response. In cervical cancer, higher densities of CD8\u0026thinsp;+\u0026thinsp;T cells within the tumor are associated with improved ICI responsiveness[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. The Immunoscore, validated by Pag\u0026egrave;s et al.[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], quantifies CD3\u0026thinsp;+\u0026thinsp;and CD8\u0026thinsp;+\u0026thinsp;T-cell density within the tumor and has proven to be a strong prognostic marker, even surpassing TNM staging system in colorectal cancer. Furthermore, the Immunoscore has shown predictive value for ICI response across multiple malignancies[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Importantly, validated data demonstrate that the Immunoscore applies to biopsy specimens as well as resected tumors, supporting its potential relevance in GAS, which is often diagnosed at advanced, unresectable stages. Although no studies have yet evaluated the Immunoscore in GAS, it may provide valuable prognostic information and guide treatment strategies.\u003c/p\u003e\u003cp\u003eGAS frequently arises in the upper cervix and is often diagnosed at advanced stages[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], making it difficult to obtain adequate tumor tissue for comprehensive analysis. Histopathological evaluation based solely on biopsy specimens has limitations, as it may not fully capture tumor heterogeneity. Imaging-based evaluations may therefore contribute to a more complete assessment. In this study, we aimed to develop a predictive model for the TME in GAS using imaging analysis.\u003c/p\u003e\u003cp\u003eRadiomics, a technique that extracts and analyzes a large number of quantitative features from medical imaging modalities such as computed tomography or magnetic resonance imaging (MRI), has gained increasing use in oncology. It has been reported to predict malignancy, histological subtypes, driver gene mutations, treatment response, and prognosis across multiple cancer types[\u003cspan additionalcitationids=\"CR12 CR13 CR14\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. In endometrial cancer, for example, MRI-based radiomic features have been shown to predict myometrial invasion and lymphovascular space invasion (LVSI)[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. In cervical cancer, MRI radiomics has been applied to predict response to concurrent chemoradiotherapy[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The objective of our study was to explore the potential of MRI-based radiomics in predicting the immune status of GAS. We characterized the TME of GAS in comparison with UEA and further performed radiomics analysis focused on GAS, a setting in which tissue-based evaluation is often limited and imaging-based approaches may have particular clinical value.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003ePatient selection\u003c/h2\u003e\u003cp\u003ePatients with a pathological diagnosis of GAS who were treated at our institution between December 2018 and February 2023 were enrolled in this study. The inclusion criteria were: (1) clinical stage greater than stage IB1 (FIGO 2018), (2) presence of an identifiable tumor on pre-treatment MRI, and (3) no prior chemotherapy, immunotherapy, or radiotherapy before diagnosis. Patients who underwent pre-treatment MRI at other hospitals were also included. A total of 16 patients met the eligibility criteria and were included in the analysis. Descriptive statistics were calculated for all study variables. For comparison of T-cell infiltration, 14 patients with UEA were selected as controls. To minimize potential confounding from differences in tumor progression, UEA patients were manually matched to the GAS cohort by tumor size, depth of invasion, and clinical stage distribution. Collected clinical variables included age, year of diagnosis, clinical stage (FIGO 2018), tumor size, depth of stromal invasion, LVSI, paracervical invasion, vaginal wall invasion, lymph node metastasis, distant metastasis, and treatment received. Survival outcomes included follow-up duration after surgery, recurrence status, and vital status at the last follow-up. Tumor morphology on MRI was classified according to the proportion of cystic components: tumors with \u0026ge;\u0026thinsp;50% cystic area were categorized as cystic-dominant, and those with \u0026lt;\u0026thinsp;50% as solid-dominant.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eImmunohistochemistry, TIL quantification, and calculation of Immunoscore\u003c/h3\u003e\n\u003cp\u003eArchived tissue samples from 16 patients with GAS and 14 with UEA, collected during routine clinical care, were used for analysis. Formalin-fixed, paraffin-embedded tissue blocks were serially sectioned at 4 \u0026micro;m. Sections were deparaffinized, rehydrated, and subjected to antigen retrieval by heating in a water bath at 95\u0026deg;C for 20 minutes. After blocking endogenous peroxidase activity, the sections were stained with antibodies against CD3 (790\u0026ndash;4341, Roche), CD4 (790\u0026ndash;4423, Roche), CD8 (M7103, Dako), and FoxP3 (ab20034, Abcam). They were incubated overnight at 4\u0026deg;C with monoclonal primary antibody, followed by amplification using a labeled polymer/HRP system (pv-6001, Zhongshan Golden Bridge Biotechnologies) at 37\u0026deg;C for one hour. Visualization was achieved with 3,3'-diaminobenzidine (DAB), and counterstaining was performed with hematoxylin. Digital images of stained slides were scanned at 40\u0026times; magnification with a resolution of 0.2 \u0026micro;m/pixel.\u003c/p\u003e\u003cp\u003eStrataQuest version 7 software (TissueGnostics) was used to quantify CD3 and CD8 expression. Tumor regions were manually segmented, and CD3\u0026thinsp;+\u0026thinsp;and CD8\u0026thinsp;+\u0026thinsp;cells were quantified based on DAB-positive nuclei. Thresholding was applied to ensure accurate detection for each core, and results were reported as cells/mm\u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe Immunoscore methodology established by Pag\u0026egrave;s et al.[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] was adapted for this study. CD3\u0026thinsp;+\u0026thinsp;and CD8\u0026thinsp;+\u0026thinsp;cell densities in the tumor region were compared across all patients and converted to percentiles. Given that GAS frequently lack a discrete mass and often exhibit infiltrative growth with indistinct margins, and that some cases were represented only by biopsy specimens, differentiation between tumor center and invasive margin was not consistently feasible. Therefore, T-cell infiltration was assessed across the entire tumor region. The mean of the CD3\u0026thinsp;+\u0026thinsp;and CD8\u0026thinsp;+\u0026thinsp;percentiles was used to classify patients into two categories: low (\u0026lt;\u0026thinsp;25%), intermediate-to-high (25\u0026ndash;100%). Progression-free survival (PFS) was then compared between Immunoscore groups among patients with GAS.\u003c/p\u003e\n\u003ch3\u003eRadiomic feature extraction\u003c/h3\u003e\n\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the workflow of radiomics analysis. Three-dimensional radiomic features were extracted from the primary tumor (PT) and the surrounding peritumoral region (PR) using axial and sagittal T2-weighted images (T2WIs) and axial diffusion-weighted imaging (DWI) obtained at diagnosis. The PT was delineated by a radiation oncologist (Y.S.) with over 20 years of experience. To characterize the PR, 10 volumes of interest (VOIs) were generated by isotropically expanding the tumor margin by 2, 4, 6, 8, and 10 mm, with each expansion defining a peritumoral volume (i.e., the margin excluding the tumor itself). All MRI images underwent preprocessing, including intensity standardization, resizing, and intensity binning (see the supplementary material). For each combination of MRI sequence (three types) and VOI (11 types), 1,302 radiomic features were extracted using PyRadiomics version 3.0.28. These included 18 first-order features, 75 texture features from the original images, and 1,209 filtered features from 13 image transformations (eight wavelet decompositions combining high- and low-pass filters along the x, y, and z axes, and five Laplacian of Gaussian filters with σ values of 1, 2, 3, 4, and 5 mm).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eTo evaluate interobserver reproducibility, a second radiation oncologist (M.S.) with over 10 years of experience, blinded to the initial delineations, independently delineated the PT. Radiomic features were extracted using the same procedure. Features with an intraclass correlation coefficient (ICC)\u0026thinsp;\u0026gt;\u0026thinsp;0.75 were considered reproducible and included in subsequent analyses.\u003c/p\u003e\n\u003ch3\u003eRadiomic feature selection associated with tumor immune microenvironment\u003c/h3\u003e\n\u003cp\u003eThe Immunoscore, validated by Pag\u0026egrave;s et al.,[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] is a percentile-based metric that stratifies immune cell densities into categorical scores. However, this transformation introduces non-linearity, making it less suitable as a target variable in linear regression models due to potential instability and reduced reproducibility. To address this limitation, we introduced a continuous alternative\u0026mdash;the TIL density index\u0026mdash;defined as the mean density of CD3\u0026thinsp;+\u0026thinsp;and CD8\u0026thinsp;+\u0026thinsp;cells (cells/mm\u0026sup2;). This index was used as a surrogate for immune status, and radiomic models were developed to predict its value.\u003c/p\u003e\u003cp\u003eFeature preselection was performed as follows: (1) the log-transformed TIL density index was used as the target variable to improve normality and ensure linearity for regression analysis. (2) All radiomic features were standardized using Z-score normalization. (3) Features showing a Pearson correlation of (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left|{r}_{\\text{y}}\\right|\u0026gt;0.5\\)\u003c/span\u003e\u003c/span\u003e) with the target variable were selected as candidate predictors. (4) To reduce redundancy, Pearson correlation coefficients were calculated between all candidate features. If two features had a very strong inter-feature correlation (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left|{r}_{\\text{f}}\\right|\u0026gt;0.9\\)\u003c/span\u003e\u003c/span\u003e), the feature with the lower absolute correlation with the target variable (lower \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left|{r}_{\\text{y}}\\right|\\)\u003c/span\u003e\u003c/span\u003e) was excluded.\u003c/p\u003e\n\u003ch3\u003eRadiomic feature map visualization\u003c/h3\u003e\n\u003cp\u003eTo examine the spatial distribution of radiomic features and improve interpretability, feature maps were generated using the feature with the highest reproducibility and strongest correlation with the log-transformed TIL density index. These maps were visually assessed to identify tumor regions contributing most to prediction, and findings were compared with histopathological evaluations of cell distribution in the tumor center, margin, and adjacent non-tumor areas.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eDeveloping a radiomics-based prediction model of tumor immune microenvironment\u003c/h2\u003e\u003cp\u003eFor clustering, an unsupervised model was constructed by applying Principal component analysis (PCA) to reduce the preselected radiomic features to two principal components, followed by K-means clustering to stratify patients according to immune status (low vs. intermediate\u0026ndash;high).\u003c/p\u003e\u003cp\u003eFor regression, a multivariable linear model was developed to predict the log-transformed TIL density index, using features further selected from the preselected set by the least absolute shrinkage and selection operator (LASSO). The LASSO regularization parameter (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\lambda\\:\\)\u003c/span\u003e\u003c/span\u003e) was optimized via grid search (Supplementary Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eStatistical analyses of patient characteristics, outcomes, and TIL infiltration were performed using SPSS (version 29.0; SPSS Inc., Chicago, IL, USA). Statistical significance was set at P\u0026thinsp;\u0026lt;\u0026thinsp;.05. Continuous variables are expressed as medians, with group comparisons performed using the Mann\u0026ndash;Whitney U test. Categorical variables were analyzed using the χ\u003csup\u003e2\u003c/sup\u003e test or Fisher\u0026rsquo;s exact test. PFS was analyzed with the Kaplan\u0026ndash;Meier method and log-rank test. Graphs were generated using GraphPad Prism (version 10.6.0).\u003c/p\u003e\u003cp\u003eStatistical analyses for feature selection and model development were performed using Python (version 3.9.7) and R (version 4.4.2). Interobserver reproducibility was assessed by ICC with the \u003cem\u003eirr\u003c/em\u003e package (version 0.84.1) in R. PCA and K-means clustering were conducted with the decomposition and cluster modules of scikit-learn (version 1.6.1) in Python. The clustering model was evaluated by accuracy, precision, recall, specificity, and F1 score. LASSO regression was performed using the \u003cem\u003elinear_model\u003c/em\u003e module of scikit-learn. Regression model performance was assessed by the Pearson correlation coefficient, coefficient of determination (R\u0026sup2;), and mean absolute error (MAE). Multicollinearity was considered acceptable when the variance inflation factor (VIF) was \u0026lt;\u0026thinsp;10, calculated with the \u003cem\u003estats\u003c/em\u003e module of SciPy (version 1.10.1) in Python.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eEthical declarations\u003c/h3\u003e\n\u003cp\u003e This retrospective study was approved by the Institutional Review Board (IRB) of Keio University Hospital (No. 20030107 and 20110275), with informed consent waived due to its retrospective design. Use of tissue samples was approved by the IRB, and an opt-out consent process was applied in accordance with institutional and national ethical regulations. All methods were carried out in accordance with the relevant guidelines and regulations.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003ePatient characteristics\u003c/h2\u003e\u003cp\u003eOf the 16 patients with GAS included in this study, the median age at diagnosis was 52 years (range, 33\u0026ndash;70 years). Six patients (37.5%) were diagnosed at Stage IB, while the remaining 10 (62.5%) were at Stage II or higher (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The median tumor size was 31 mm (range, 9\u0026ndash;60 mm), and the median depth of stromal invasion was 11 mm (range, 1\u0026ndash;25 mm). The median follow-up duration was 22 months (range, 8\u0026ndash;59 months). Of the 16 patients, 10 (62.5%) underwent surgical treatment, 4 (25.0%) received concurrent chemoradiotherapy, and 2 (12.5%) received chemotherapy alone. Among the 14 patients with UEA, the median age at diagnosis was 38 years (range, 29\u0026ndash;52 years). The overall clinicopathological characteristics were comparable between the two groups, with no significant differences observed in tumor size or depth of invasion. GAS cases exhibited variable MRI morphologies, with 7 of 16 (43.8%) showing cystic-dominant features (\u0026ge;\u0026thinsp;50% cystic component) and 9 of 16 (56.2%) showing solid-dominant features. In contrast, all UEA cases (14/14) demonstrated solid morphology.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePatient characteristics:\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGAS\u003c/p\u003e\u003cp\u003en\u0026thinsp;=\u0026thinsp;16\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eUEA\u003c/p\u003e\u003cp\u003en\u0026thinsp;=\u0026thinsp;14\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ep value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eAge (years), median (range)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e52 (33\u0026ndash;70)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e38 (29\u0026ndash;52)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.001\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eStage (FIGO2018)\u003c/p\u003e\u003cp\u003e(n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIB\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6 (37.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e9 (64.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e0.272\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eII-IV\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e10 (62.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e5 (35.7)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eLVSI (n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5/10 (50.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e10/12 (83.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.172\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eTumor size (mm), median (range)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e31 (9\u0026ndash;60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e25 (17\u0026ndash;55)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.3179\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eDepth of invasion (mm), median (range)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e11 (1\u0026ndash;25)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e10 (5\u0026ndash;27)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.000\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eLymph node metastasis (n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5 (31.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4 (28.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.000\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eParametrial invasion (n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e8 (50.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2 (14.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.058\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eVaginal invasion (n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4 (25.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3 (21.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.000\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eLesion morphology (n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eCystic-dominant\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7 (43.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0 (0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e0.007\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eSolid-dominant\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9 (56.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e14 (100)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eRecurrence (n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9 (56.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2 (14.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.026\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eDeath (n, %)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6 (37.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1 (7.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.086\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eObservation period (months), median (range)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e22 (8\u0026ndash;59)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e34 (10\u0026ndash;50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.400\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003eGAS; gastric-type adenocarcinoma, UEA; usual endocervical adenocarcinoma, LVSI; lymphvascular space invasion\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003ea; Mann-Whitney U test, b; chi-square test, c; Fischer\u0026rsquo;s exact test\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eTumor immune microenvironment in GAS\u003c/h2\u003e\u003cp\u003eImmunohistochemical analysis was performed to evaluate TILs, with results shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Representative cases of early-stage and advanced-stage GAS and UEA are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea. A clear difference in TIL density between GAS and UEA was observed. GAS tumors exhibited lower infiltration of CD3\u0026thinsp;+\u0026thinsp;and CD8\u0026thinsp;+\u0026thinsp;T cells, and FoxP3-positive cells were also fewer in GAS (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea, b).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eGAS cases showed a trend toward a higher frequency of Immunoscore-low tumors (37.5% vs. 7.1%, P\u0026thinsp;=\u0026thinsp;0.086) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec). Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed compares immune status by clinical stage in GAS and UEA, assessed using the TIL density index. GAS tumors exhibited a distinct pattern, with markedly reduced T-cell infiltration, particularly in early-stage disease (TIL density index: 243 cells/mm\u0026sup2; in early-stage vs. 875 cells/mm\u0026sup2; in advanced stage; P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). By contrast, UEA cases demonstrated T-cell infiltration from early stages, with no significant variation across clinical stages. Among GAS patients, those with a high Immunoscore had significantly poorer prognosis compared with those with a low Immunoscore (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ee).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eImmunoscore-associated radiomic features\u003c/h2\u003e\u003cp\u003eA total of 12 radiomic features were selected through the preselection process as candidate predictors of immune status (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The number of features selected at each step, stratified by image type and VOI, is summarized in Supplementary Fig. S2. \u003cem\u003eRunEntropy\u003c/em\u003e from the gray-level run-length matrix (GLRLM), extracted from the PT volume with a 10-mm peritumoral margin on the original axial DWI sequence (\u003cem\u003eAxialDWI_PT\u0026thinsp;+\u0026thinsp;PR10mm_Original_GLRLM_RunEntropy\u003c/em\u003e), showed high reproducibility (ICC\u0026thinsp;=\u0026thinsp;0.773) and the strongest correlation with the log-transformed TIL density index (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{r}_{\\text{y}}\\)\u003c/span\u003e\u003c/span\u003e = 0.69, P\u0026thinsp;=\u0026thinsp;0.003). \u003cem\u003eRunEntropy\u003c/em\u003e quantifies the uncertainty or randomness in the distribution of run lengths and gray levels within the VOI; higher values indicate greater textural heterogeneity. This feature also showed consistent reproducibility and predictive performance when extracted from the PT volume with an additional peritumoral margin of 4\u0026ndash;10 mm (Supplementary Fig. S3). However, when evaluated using the peritumoral volume alone, the feature did not meet reproducibility thresholds and was excluded.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eRadiomic features associated with tumor immune microenvironment\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModality\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVolume of interest\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eImage filter\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFeature category\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eFeature name\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003er\u003csub\u003ey\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003ep value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial DWI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u0026thinsp;+\u0026thinsp;PR10mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOriginal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLRLM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eRunEntropy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.69\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.003\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial DWI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u0026thinsp;+\u0026thinsp;PR6mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoG (σ\u0026thinsp;=\u0026thinsp;1mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFirstorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMean\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-0.65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.007\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial T2WI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoG (σ\u0026thinsp;=\u0026thinsp;2mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLCM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eInverseDifferenceNormalized\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.59\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.015\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial T2WI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u0026thinsp;+\u0026thinsp;PR10mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoG (σ\u0026thinsp;=\u0026thinsp;4mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLSZM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSmallAreaLowGrayLevelEmphasis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-0.58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.019\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial DWI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u0026thinsp;+\u0026thinsp;PR8mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOriginal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLCM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eInformationalMeasureOfCorrelation2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.57\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.021\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSagittal T2WI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePR4mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ewavelet-HLL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFirstorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSkewness\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.56\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.025\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial DWI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u0026thinsp;+\u0026thinsp;PR2mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoG (σ\u0026thinsp;=\u0026thinsp;1mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLCM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMaximalCorrelationCoefficient\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.55\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.028\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSagittal T2WI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePR2mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoG (σ\u0026thinsp;=\u0026thinsp;1mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLDM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSmallDependenceLowGrayLevelEmphasis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.53\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.036\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial T2WI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ewavelet-LHL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFirstorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMedian\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.52\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.041\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSagittal T2WI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u0026thinsp;+\u0026thinsp;PR4mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ewavelet-HLL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLRLM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eLongRunHighGrayLevelEmphasis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-0.52\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.041\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial DWI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u0026thinsp;+\u0026thinsp;PR2mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ewavelet-LHL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLSZM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eHighGrayLevelZoneEmphasis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.043\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial DWI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOriginal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNGTDM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eStrength\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.044\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"7\"\u003er\u003csub\u003ey\u003c/sub\u003e = Pearson\u0026rsquo;s correlation coefficient with the log-transformed average density of CD3-positive and CD8-positive cells; DWI\u0026thinsp;=\u0026thinsp;Diffusion-Weighted Imaging; T2WI\u0026thinsp;=\u0026thinsp;T2-Weighted Imaging; PT\u0026thinsp;=\u0026thinsp;Primary Tumor; PRxmm\u0026thinsp;=\u0026thinsp;peritumoral region with an isotropic margin of x mm; LoG-σxmm\u0026thinsp;=\u0026thinsp;Laplacian of Gaussian filter with σ\u0026thinsp;=\u0026thinsp;x mm; wavelet-XYZ\u0026thinsp;=\u0026thinsp;wavelet-decomposed image in which each letter (X, Y, Z) denotes low-pass (L) or high-pass (H) filtering applied along the corresponding image axis; GLCM\u0026thinsp;=\u0026thinsp;Gray-Level Co-occurrence Matrix; GLRLM\u0026thinsp;=\u0026thinsp;Gray-Level Run-Length Matrix; GLSZM\u0026thinsp;=\u0026thinsp;Gray-Level Size-Zone Matrix; GLDM\u0026thinsp;=\u0026thinsp;Gray-Level Dependence Matrix; NGTDM\u0026thinsp;=\u0026thinsp;Neighborhood Gray-Tone Difference Matrix.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eRadiomic feature mapping\u003c/h2\u003e\u003cp\u003eWe generated feature maps of \u003cem\u003eDWIax_GTV\u0026thinsp;+\u0026thinsp;10mm_original_GLRLM_RunEntropy\u003c/em\u003e, one of the most reproducible radiomic features and the one most strongly correlated with TIL density index. Representative MRI images, corresponding radiomic feature maps, and histologic cell density are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Cases 1 and 2 represent tumors with cystic morphology. The cystic regions in Case 1 demonstrated high signal intensity on DWI, whereas those in Case 2 appeared DWI-low. These differences were reflected in the \u003cem\u003eRunEntropy\u003c/em\u003e values: Case 1 showed high and heterogeneous \u003cem\u003eRunEntropy\u003c/em\u003e within the cystic regions, whereas Case 2 exhibited uniformly low values. Cases 3 and 4 represent tumors with solid morphology. In these cases, regions with high \u003cem\u003eRunEntropy\u003c/em\u003e values were typically located at the tumor margins rather than the center (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea). Histopathological analysis confirmed that total cell density was highest in the tumor center, decreased at the tumor margin, and was lowest in non-tumor tissue, with marked heterogeneity at the tumor margins. The distribution of CD3\u0026thinsp;+\u0026thinsp;T-cell density also aligned with the regions of high total cell density, as shown in the heatmap (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb\u0026ndash;e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003eRadiomics-based prediction of tumor immune microenvironment\u003c/h2\u003e\u003cp\u003eThe performance of the clustering approach is summarized in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea\u0026ndash;c. The clustering model achieved an accuracy of 0.875, precision of 0.750, recall of 1.000, specificity of 0.800, and an F1 score of 0.857, with two false-negative cases misclassified. PCA of the 12 preselected features revealed that the first principal component (PC1) was positively correlated with the log-transformed TIL density index (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{r}_{\\text{y}}\\)\u003c/span\u003e\u003c/span\u003e = 0.86, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The contribution of \u003cem\u003eDWIax_GTV\u0026thinsp;+\u0026thinsp;10mm_original_GLRLM_RunEntropy\u003c/em\u003e accounted for 11.4% of the variance in PC1, the largest among the 12 features, making it the most influential radiomic feature for PC1 (see Supplementary Fig. S4).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe performance of the regression model is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed. Multivariable linear regression demonstrated strong predictive performance, with a Pearson correlation coefficient of 0.93 (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) between predicted and observed values of the log-transformed TIL density index, an R\u0026sup2; of 0.86, and a MAE of 0.12. Eight radiomic features were selected from the 12 preselected features by LASSO regression and are summarized in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e. Although all selected features showed statistically significant associations in univariate analysis, none remained significant in the multivariable model, likely due to the limited sample size. All VIFs were \u0026lt;\u0026thinsp;10, indicating minimal multicollinearity.\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eComparison with UEA revealed that GAS exhibits a distinctly different TME, characterized by sparse lymphocytic infiltration and an immunologically \"cold\" phenotype. Interestingly, in GAS, more advanced tumors tend to display increased immune activity compared with early-stage cases, suggesting potential susceptibility to immunotherapy. The limited immunogenicity of early GAS, likely due to highly differentiated morphology with minimal atypia and weak antigenicity, may hinder the initiation of immune responses and contribute to tumor invasion and progression.\u003c/p\u003e\u003cp\u003eIn other cancer types, immunologically \"hot\" tumors are often associated with better prognosis[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], which contrasts with the trend observed in GAS. This discrepancy may reflect the tendency for low TIL infiltration in early-stage disease and high TIL infiltration in advanced stages.\u003c/p\u003e\u003cp\u003eIn this study, we analyzed TIL infiltration patterns in GAS and established an MRI-based radiomics model to predict immune status. These results indicate that radiomics has potential as a noninvasive approach for assessing the TME in GAS.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eTo validate the reliability of the radiomics prediction model and enhance interpretability, we examined the correspondence between extracted radiomic features and underlying histopathological architecture. \u003cem\u003eRunEntropy\u003c/em\u003e, the most reproducible and TIL density-correlated radiomic feature in our study, is a texture feature derived from the gray-level run length matrix. It quantifies variability in the lengths of consecutive runs of pixels with the same intensity. It measures the randomness or complexity of these run lengths, with higher values indicating greater heterogeneity within the tissue. Histopathologically, increased \u003cem\u003eRunEntropy\u003c/em\u003e has been associated with diverse tumor microarchitectures, including variations in cellular density, glandular structures, and cystic or necrotic areas.\u003c/p\u003e\u003cp\u003eGAS can exhibit at least two distinct morphological patterns: (1) tumors predominantly composed of macrocystic structures, and (2) tumors mainly consisting of solid components. Our findings highlight that radiomic features derived from both cystic and solid components provide important information for predicting the Immunoscore. In cystic-dominant tumors, variations in DWI signal within the cystic components (Cases 1 and 2 in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea) likely reflect differences in tumor-secreted mucin profiles. Low-viscosity fluid with low protein content typically appears hypointense on DWI, whereas hemorrhagic or protein-rich mucin tends to appear hyperintense. Supporting this, prior reports have shown that during progression from lobular endocervical glandular hyperplasia to GAS, the positivity rates of MUC6, a marker of gastric-type mucin, and αGlcNAc, a carbohydrate modification of gastric mucin, decrease[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. This suggests that alterations in mucin composition accompany tumor progression and may influence the immune microenvironment. In such tumors, radiomic features reflecting signal intensity or internal heterogeneity on DWI may correspond to cystic spaces containing serous or mucinous fluid of varying viscosity and protein content. These fluid characteristics could underlie the observed signal differences and may contribute to an immunosuppressive TME.\u003c/p\u003e\u003cp\u003eIn contrast, solid-dominant tumors, as shown in Cases 3 and 4 in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea, were characterized by higher \u003cem\u003eRunEntropy\u003c/em\u003e values at the tumor margin, which corresponded histologically to regions with heterogeneous cell density and irregular lymphocyte infiltration. CD3 immunostaining confirmed that tumor centers often exhibited uniform, dense infiltration, whereas the periphery demonstrated more variable immune cell distribution. These findings suggest that radiomic features extracted from tumor margins\u0026mdash;where signal heterogeneity and irregular interfaces are frequently observed\u0026mdash;may reflect stromal reaction, lymphocytic infiltration, or tumor\u0026ndash;stroma interactions, all critical components of the local immune contexture.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eStrengths of this study include the use of routinely acquired MRI scans without reliance on contrast-enhanced sequences or PET-MRI, allowing broad applicability across clinical settings and imaging protocols. GAS frequently exhibits poorly defined margins and adjacent organ invasion, which can compromise segmentation reproducibility and reduce the accuracy of radiomics-based models. To address this, we performed manual segmentation by two experienced radiologists and extracted only highly reproducible features, enhancing the robustness of our analysis. Another strength lies in the noninvasive, three-dimensional nature of MRI-based radiomics, which allows comprehensive assessment of the entire tumor\u0026mdash;a particular advantage in GAS, where tissue sampling is often limited. With continued accumulation of clinical and imaging data, this approach holds promise for improving preoperative assessment, predicting therapeutic responses, and guiding personalized treatment strategies in GAS.\u003c/p\u003e\u003cp\u003eSince TME is heterogeneous within tumors, accurate pathological assessment of the Immunoscore ideally requires evaluation of the entire resected specimen. However, prior studies of Immunoscore in colorectal cancer have included a substantial number of endoscopic biopsy samples, permitting evaluation on limited tissue[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Similarly, in GAS, many advanced cases are not amenable to resection. Therefore, consistent with prior reports, we included Immunoscore assessment based on biopsy samples. This may have introduced some uncertainty in Immunoscore evaluation in advanced compared with early-stage cases.\u003c/p\u003e\u003cp\u003eAnother limitation is the potential risk of overfitting in radiomics modeling due to the limited sample size. To mitigate this risk, we employed strategies, including feature preselection, reproducibility assessment using ICC, correlation filtering, and LASSO regularization. While these approaches reduce model complexity and improve robustness, external validation with larger, multi-institutional cohorts will be essential to confirm generalizability.\u003c/p\u003e\u003cp\u003eAlthough the Immunoscore shows promise as a predictive biomarker for immunotherapy responsiveness, immune landscapes vary across organs and cancer types. An absolute cutoff for Immunoscore has not been established for cervical cancer. Further investigations are needed to determine appropriate thresholds, validate prognostic and predictive relevance, and assess clinical applicability in decision-making for patients with cervical cancer.\u003c/p\u003e\u003cp\u003eIn this study, we characterized the T-cell infiltration profile of GAS and successfully developed an MRI-based radiomics model capable of predicting immune status. Our findings suggest that radiomics may serve as a promising noninvasive tool for evaluating the TME in GAS.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003e\u003cb\u003eAdditional Information\u003c/b\u003e\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis study was supported by JSPS KAKENHI Grant Number JP25K02782 (Grant-in-Aid for Scientific Research (B)) and JP25K10545 (Grant-in-Aid for Scientific Research (C)).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eRisa Matsuda and Hiroshi Nishio contributed to study concept and design; Risa Matsuda and Kohei Oguma drafted the manuscript; Hiroshi Nishio obtained funding and supervised the study; Risa Matsuda and Hiroshi Nishio collected and analyzed clinical data; Kohei Oguma performed radiomics analysis; Miyuki Saito conducted immunohistochemistry and automated cell quantification; Maho Kurihara, Masafumi Sawada, and Yutaka Shiraishi contributed to radiodiagnosis and segmentation; Yutaka Shiraishi and Hiroshi Nishio provided senior supervision; Masaki Sugawara, Tomoya Matsui, and Takashi Iwata contributed to data interpretation and critical manuscript review; Masahiro Jinzaki, Atsuya Takeda, and Wataru Yamagami provided expert guidance as departmental chairs; all authors reviewed and approved the final version of the manuscript. Risa Matsuda and Kohei Oguma contributed equally to this study.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eKusanagi, Y. et al. Absence of high-risk human papillomavirus (HPV) detection in endocervical adenocarcinoma with gastric morphology and phenotype. \u003cem\u003eAm. J. Pathol.\u003c/em\u003e \u003cb\u003e177\u003c/b\u003e, 2169\u0026ndash;2175. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2353/ajpath.2010.100323\u003c/span\u003e\u003cspan address=\"10.2353/ajpath.2010.100323\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMikami, Y. Gastric-type mucinous carcinoma of the cervix and its precursors - historical overview. \u003cem\u003eHistopathology\u003c/em\u003e \u003cb\u003e76\u003c/b\u003e, 102\u0026ndash;111. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/his.13993\u003c/span\u003e\u003cspan address=\"10.1111/his.13993\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStolnicu, S. et al. Diagnostic Algorithmic Proposal Based on Comprehensive Immunohistochemical Evaluation of 297 Invasive Endocervical Adenocarcinomas. \u003cem\u003eAm. J. Surg. Pathol.\u003c/em\u003e \u003cb\u003e42\u003c/b\u003e, 989\u0026ndash;1000. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/PAS.0000000000001090\u003c/span\u003e\u003cspan address=\"10.1097/PAS.0000000000001090\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNishio, S. et al. Analysis of gastric-type mucinous carcinoma of the uterine cervix - An aggressive tumor with a poor prognosis: A multi-institutional study. \u003cem\u003eGynecol. Oncol.\u003c/em\u003e \u003cb\u003e153\u003c/b\u003e, 13\u0026ndash;19. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ygyno.2019.01.022\u003c/span\u003e\u003cspan address=\"10.1016/j.ygyno.2019.01.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKojima, A. et al. Gastric morphology and immunophenotype predict poor outcome in mucinous adenocarcinoma of the uterine cervix. \u003cem\u003eAm. J. Surg. Pathol.\u003c/em\u003e \u003cb\u003e31\u003c/b\u003e, 664\u0026ndash;672. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/01.pas.0000213434.91868.b0\u003c/span\u003e\u003cspan address=\"10.1097/01.pas.0000213434.91868.b0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2007).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTewari, K. S. et al. Pembrolizumab or Placebo Plus Chemotherapy With or Without Bevacizumab for Persistent, Recurrent, or Metastatic Cervical Cancer: Subgroup Analyses From the KEYNOTE-826 Randomized Clinical Trial. \u003cem\u003eJAMA Oncol.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 185\u0026ndash;192. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jamaoncol.2023.5410\u003c/span\u003e\u003cspan address=\"10.1001/jamaoncol.2023.5410\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, Y. et al. Multiparametric immune profiling of advanced cervical cancer to predict response to programmed death-1 inhibitor combination therapy: an exploratory study of the CLAP trial. \u003cem\u003eClin. Transl Oncol.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e, 256\u0026ndash;268. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s12094-022-02945-1\u003c/span\u003e\u003cspan address=\"10.1007/s12094-022-02945-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePages, F. et al. International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e391\u003c/b\u003e, 2128\u0026ndash;2139. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0140-6736(18)30789-X\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(18)30789-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEljilany, I. et al. The T Cell Immunoscore as a Reference for Biomarker Development Utilizing Real-World Data from Patients with Advanced Malignancies Treated with Immune Checkpoint Inhibitors. \u003cem\u003eCancers (Basel)\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/cancers15204913\u003c/span\u003e\u003cspan address=\"10.3390/cancers15204913\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKido, A. et al. Magnetic resonance appearance of gastric-type adenocarcinoma of the uterine cervix in comparison with that of usual-type endocervical adenocarcinoma: a pitfall of newly described unusual subtype of endocervical adenocarcinoma. \u003cem\u003eInt. J. Gynecol. Cancer\u003c/em\u003e. \u003cb\u003e24\u003c/b\u003e, 1474\u0026ndash;1479. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/IGC.0000000000000229\u003c/span\u003e\u003cspan address=\"10.1097/IGC.0000000000000229\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGevaert, O. et al. Predictive radiogenomics modeling of EGFR mutation status in lung cancer. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 41674. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/srep41674\u003c/span\u003e\u003cspan address=\"10.1038/srep41674\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePesapane, F. et al. How Radiomics Can Improve Breast Cancer Diagnosis and Treatment. \u003cem\u003eJ. Clin. Med.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/jcm12041372\u003c/span\u003e\u003cspan address=\"10.3390/jcm12041372\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSon, J., Lee, S. E., Kim, E. K. \u0026amp; Kim, S. Prediction of breast cancer molecular subtypes using radiomics signatures of synthetic mammography from digital breast tomosynthesis. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 21566. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-020-78681-9\u003c/span\u003e\u003cspan address=\"10.1038/s41598-020-78681-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTagliafico, A. S. et al. Overview of radiomics in breast cancer diagnosis and prognostication. \u003cem\u003eBreast\u003c/em\u003e \u003cb\u003e49\u003c/b\u003e, 74\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.breast.2019.10.018\u003c/span\u003e\u003cspan address=\"10.1016/j.breast.2019.10.018\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu, Y. J., Wu, F. Z., Yang, S. C., Tang, E. K. \u0026amp; Liang, C. H. Radiomics in Early Lung Cancer Diagnosis: From Diagnosis to Clinical Decision Support and Education. \u003cem\u003eDiagnostics (Basel)\u003c/em\u003e. 12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics12051064\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics12051064\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHan, Y. et al. Predicting myometrial invasion in endometrial cancer based on whole-uterine magnetic resonance radiomics. \u003cem\u003eJ. Cancer Res. Ther.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e, 1648\u0026ndash;1655. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.4103/jcrt.JCRT_1393_20\u003c/span\u003e\u003cspan address=\"10.4103/jcrt.JCRT_1393_20\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa, W. et al. Predictive value of models based on MRI radiomics and clinical indicators for lymphovascular space invasion in endometrial cancer. \u003cem\u003eBMC Cancer\u003c/em\u003e. \u003cb\u003e25\u003c/b\u003e, 796. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12885-025-14217-6\u003c/span\u003e\u003cspan address=\"10.1186/s12885-025-14217-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCai, C. et al. Longitudinal dynamic MRI radiomic models for early prediction of prognosis in locally advanced cervical cancer treated with concurrent chemoradiotherapy. \u003cem\u003eRadiat. Oncol.\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e, 181. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13014-024-02574-8\u003c/span\u003e\u003cspan address=\"10.1186/s13014-024-02574-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYun, S. et al. Immunoscore is a strong predictor of survival in the prognosis of stage II/III gastric cancer patients following 5-FU-based adjuvant chemotherapy. \u003cem\u003eCancer Immunol. Immunother\u003c/em\u003e. \u003cb\u003e70\u003c/b\u003e, 431\u0026ndash;441. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00262-020-02694-6\u003c/span\u003e\u003cspan address=\"10.1007/s00262-020-02694-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZeng, L. et al. Clinical Significance of a CD3/CD8-Based Immunoscore in Neuroblastoma Patients Using Digital Pathology. \u003cem\u003eFront. Immunol.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 878457. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fimmu.2022.878457\u003c/span\u003e\u003cspan address=\"10.3389/fimmu.2022.878457\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchoumacher, C. et al. CD3-CD8 immune score associated with a clinical score stratifies PDAC prognosis regardless of adjuvant or neoadjuvant chemotherapy. \u003cem\u003eOncoimmunology\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 2294563. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/2162402X.2023.2294563\u003c/span\u003e\u003cspan address=\"10.1080/2162402X.2023.2294563\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYamanoi, K., Ishii, K., Tsukamoto, M., Asaka, S. \u0026amp; Nakayama, J. Gastric gland mucin-specific O-glycan expression decreases as tumor cells progress from lobular endocervical gland hyperplasia to cervical mucinous carcinoma, gastric type. \u003cem\u003eVirchows Arch.\u003c/em\u003e \u003cb\u003e473\u003c/b\u003e, 305\u0026ndash;311. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00428-018-2381-6\u003c/span\u003e\u003cspan address=\"10.1007/s00428-018-2381-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"cervical cancer, gastric-type adenocarcinoma, radiomics, magnetic resonance imaging, imaging analysis, immune microenvironment","lastPublishedDoi":"10.21203/rs.3.rs-7895813/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7895813/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGastric-type adenocarcinoma (GAS) of the cervix is usually diagnosed at advanced stages and has a poor prognosis due to resistance to standard therapies. Although immune checkpoint inhibitors have improved outcomes in cervical cancer, prognosis and treatment response are strongly influenced by the tumor immune microenvironment (TME). Histopathological assessment of GAS is challenging because it often arises in the upper cervix. This study aimed to predict the TME in GAS using MRI-based radiomics. We enrolled 16 patients with GAS treated at our institution. Tumor-infiltrating lymphocytes (TILs) were evaluated by immunohistochemistry, quantifying them with the Immunoscore. Fourteen patients with usual endocervical adenocarcinoma (UEA) served as controls. A total of 1,309 radiomic features were extracted from the primary tumor and peritumoral region on pre-treatment MRI images. After feature selection, clustering, and regression models were developed to predict the TME in GAS. GAS exhibited significantly lower T-cell infiltration than UEA, particularly in early-stage tumors. The clustering model achieved 87.5% accuracy in Immunoscore classification, and the regression model showed a strong correlation with observed TIL densities (r\u0026thinsp;=\u0026thinsp;0.93, P\u0026thinsp;\u0026lt;\u0026thinsp;.001). These findings suggest that MRI-based radiomics may serve as a noninvasive biomarker for predicting the TME in GAS.\u003c/p\u003e","manuscriptTitle":"Radiomics modeling to predict the tumor immune-microenvironment of mucinous adenocarcinoma of gastric-type cervical cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-26 07:01:32","doi":"10.21203/rs.3.rs-7895813/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-02-03T13:14:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"302857912913938444321978382447089061279","date":"2026-01-29T12:21:45+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-14T06:55:48+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-10-28T13:25:43+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-21T11:56:16+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-21T11:55:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-10-19T00:41:26+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5c60d40f-fa8b-462d-aa8e-bdc6f90d8948","owner":[],"postedDate":"November 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":58482498,"name":"Health sciences/Biomarkers"},{"id":58482499,"name":"Biological sciences/Cancer"},{"id":58482500,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":58482501,"name":"Health sciences/Oncology"}],"tags":[],"updatedAt":"2025-11-26T07:01:32+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-26 07:01:32","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7895813","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7895813","identity":"rs-7895813","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}