Development and verification of a radiomics model to forecast Ki67 index and prognosis in advanced gastric tubular adenocarcinoma.

Introduction

Gastric cancer remains one of the most prevalent and lethal malignancies worldwide, with significant challenges in diagnosis, treatment, and prognosis. Early and accurate assessment of tumor biology is essential to develop appropriate treatment strategies and improve patient outcomes [ 1 , 2 ]. Among various biomarkers, the cell proliferation marker Ki67 protein is a key factor in assessing tumor aggressiveness and response to treatment. Ki67, as a nuclear protein, is expressed in the G1, S, G2 and M phases of the cell cycle, but not in the G0 phase [ 3 ]. Ki67 expression levels have been extensively employed to assess the proliferative activity of various malignant tumors [ 4 ], and its value in clinical and prognostic evaluation has been widely recognized [ 5 ], its expression levels not only provide direct indicators of tumor proliferation activity but also are crucial in aiding diagnosis, evaluating prognosis, and selecting treatment. For instance, a high Ki67 index may suggest high tumor proliferation activity, potentially necessitating more aggressive treatment strategies [ 6 ]. Conversely, patients with a low Ki67 index might consider relatively conservative treatment approaches [ 7 ]. Additionally, the Ki67 index can be used to monitor treatment effectiveness, helping physicians assess patients’ responses to specific treatment regimens, thus allowing timely adjustments to treatment strategies to improve patient survival rates and quality of life [ 8 ].

However, assessing Ki67 levels requires invasive biopsy and immunohistochemical analysis, which can be both cumbersome and uncomfortable for the patient [ 9 ]. Therefore, developing a non-invasive and reliable technique to predict Ki67 levels is of paramount importance, such a method would streamline the evaluation process and provide crucial prognostic information promptly. Utilizing non-invasive imaging methods, such as enhanced computed tomography (CT), to estimate Ki67 levels could significantly reduce the need for invasive biopsies and minimize patient discomfort. Moreover, early detection of Ki67 levels through these methods can identify patients at high risk of rapid disease progression, allowing for earlier and potentially more effective interventions.

As an innovative interdisciplinary technology, radiomics leverages the power of deep learning and machine learning algorithms to extract and analyze a vast array of quantitative features from medical images [ 10 ]. This approach has proven immensely valuable in developing predictive models for various cancers [ 11 , 12 ], as well as in predicting the response of cancer patients to neoadjuvant chemotherapy and postoperative treatments [ 13 , 14 ]. Advances in imaging technology, especially enhanced CT, have opened new avenues for non-invasive cancer assessment. Enhanced CT, which uses contrast agents to better visualize soft tissues, not only enables detailed delineation of tumor morphology but also allows the inference of biological behaviors through the analysis of tumor enhancement patterns. These technological advances not only enhance our understanding of tumor biology but also pave the way for more personalized and precise cancer therapies.

This study aims to use the power of enhanced CT to develop a predictive model to estimate Ki67 index and predict prognosis in patients with advanced gastric tubular adenocarcinoma. To achieve these objectives, we conducted a comprehensive analysis using clinical and radiomics data from 213 patients treated at our hospital. Through advanced statistical methods, we explored the relationship between Ki67 expression, radiomics scores, and survival outcomes. This investigation highlighted the potential of radiomics in providing valuable predictive insights that are crucial for personalized treatment strategies. By accurately predicting the Ki67 index and patient prognosis, our model can significantly contribute to clinical decision-making processes, enabling more tailored and effective therapeutic approaches.

Methods and materials

Patient and tissue sample collection

All patients in this study were from our hospital, TCGA and GEPIA dataset. We collected 213 patients with complete clinical and prognostic data from 2020 to 2022, and selected 192 patients for follow-up analysis and study of radiomics. The following were the inclusion criteria: [ 1 ] Patients with advanced gastric tubular adenocarcinoma [ 2 ]. Enhanced abdominal CT was performed before operation [ 3 ]. The patient underwent surgical treatment [ 4 ]. The patient did not receive neoadjuvant therapy before surgery. Specific exclusion criteria included: [ 1 ] Patients not diagnosed with gastric cancer [ 2 ]. Patients lacking essential clinical information [ 3 ]. Patients who refused or were intolerant to surgery [ 4 ]. Patients with unclear or unavailable preoperative CT images. Postoperative staging adhered to the ninth edition of the AJCC Cancer Staging Manual. The study followed patients up until November 20, 2023.

Radiomics feature extraction

All imaging data for this study were sourced from the Department of Radiology. We used preoperative contrast-enhanced CT venous phase images with a CT slice thickness of 0.6 mm. An experienced imaging physician employed 3Dslicer software (version 5.6.1) to semi-automatically delineate the regions of interest (ROI) in all CT images twice. Following this, radiomics technology was applied to extract features from these regions. The reproducibility of the extracted features was evaluated using Intra-class Correlation Coefficients (ICC), and only features with an ICC value exceeding 0.75 were selected for further analysis in our study.

Model construction and performance testing

In this study, Pyradiomics was utilized to extract a total of 1,446 radiomics features from the imaging data. The interclass correlation coefficient (ICC) for each feature was calculated to assess reliability, and features with an ICC greater than 0.75 were retained, narrowing the selection down to 415 features. These features were then standardized using a Z-score transformation to ensure uniformity in data scaling. Considering the statistical modeling principle of a 10:1 sample-to-feature ratio, we aimed to maintain no more than 20 features. We evaluated a wide range of lambda values during Least absolute shrinkage and selection operator (Lasso) regression, resulting in multiple models with varying feature counts. Each model was assessed based on multiple performance metrics including accuracy, precision, recall, specificity, F1 score, and AUC in both training and validation datasets. Notably, the model comprising 6 radiomic features demonstrated the most favorable balance between predictive performance and model simplicity. In a 7:3 ratio, the 192 patient samples were randomized to training and validation sets, ensuring robustness in model training and evaluation. Additionally, the TCIA dataset served as an external validation set to test the model’s performance. The effectiveness of the model was assessed through the Receiver Operating Characteristic (ROC) curve and its corresponding area under the curve (AUC), providing insights into its predictive accuracy. Decision curve analysis (DCA), calibration curves and Norman diagram were used to evaluate the clinical application value of the model.

Clinical data correlation analysis

In our data, the median Ki67 index is 60. To explore the relationship between clinical features and patient prognosis in gastric cancer, univariate and multivariate Cox regression analyses were employed. Initially, univariate Cox regression identified the prognostic significance of various clinical and pathological features. Given their established relevance to overall survival (OS) both in clinical practice and literature, these features were subsequently included in the multivariate analyses. The Cox regression analyses were conducted using a suite of R packages, including “autoReg”, “tidyverse”, “ggsci”, “survival”, “survminer”, and “forestplot”.

Statistical analysis

Data analysis and graphics drawing were completed with R software version 4.2.3, GraphPad Prism 8 and SPSS 26, using tools such as “ggplot”, “survival”, “survminer”, “caret”, “glmnet”, “pROC”, “rms” and “dcurves”, data analysis used a variety of statistical methods including Fisher’s exact test, two-sided Wilcoxon test, and Spearman correlation analysis. The t test or Mann-Whitney U test was used to compare continuous variables, and the chi-square test or Fisher’s exact test was used to assess categorical variables, all statistical analyzes used a two-tailed p value < 0.05 as the significance criterion.

Results

Study population characteristics data

Initially, we collected the clinical data of advanced gastric tubular adenocarcinoma patients in our hospital, detailed in Table 1 . We categorized 213 patients into groups with high and low expression based on the median Ki67 index (Ki67 median index = 60%), and analyzed the correlation of various indicators such as age, gender, Lauren’s classification, and differentiation with Ki67 index level. The analysis revealed significant associations of Her2 ( P = 0.034), T stage ( P < 0.001), and postoperative TNM stage ( P = 0.029) with Ki67 index level. However, other indicators did not show statistical significance.

In-depth examination of Ki67 expression and its implications for prognosis in gastric cancer

To further explore the relationship between Ki67 index and the prognosis of gastric cancer, we conducted a detailed analysis of Ki67 levels across various gastric cancer samples. Data from the TCGA and GEPIA database revealed that the expression level of the MKI67 gene, which encodes the Ki67 protein, was significantly higher in tumor tissues compared to normal tissues, as illustrated in Fig. 1 A-B. The pathological sections with higher Ki67 index were stained as shown in Fig. 1 C-D and those with lower Ki67 index were stained as shown in Fig. 1 E-F.

Furthermore, we employed Kaplan-Meier survival analysis to assess the correlation between Ki67 expression and gastric cancer prognosis (Fig. 1 G-H). While the TCGA and GEPIA database showed no significant association between the Ki67 gene expression and gastric cancer prognosis.

Construction of radiomics model for Ki67 index level

We further screened from 213 patients, and finally selected 192 samples for the construction of the radiomics model (Fig. 2 and Supplementary Table 1). Pyradiomics software was used to extract the features of ROI and calculate the intraclass correlation coefficient. Retaining only features with an ICC value higher than 0.75, resulting in a set of 415 radiomics features for further analysis. These features were screened by the LASSO method (Figs. 3 A-D) to optimize the selection of model features. We then randomly assigned these patients to the training set and validation set in a 7:3 ratio.

Additionally, we applied the same methodology for ROI profiling and feature extraction on selected samples from the TCIA database, as outlined in Supplementary Table 2. This external validation set helped to further assess the generalization capability and performance of our mode (Fig. 3 E-F). This approach not only enhances the rigor of the study, but also provides important evidence about the validity of the model.

Performance verification of radiomics model for Ki67 index level

Comprehensively evaluate the indicators and AUC of the model, six essential radiomics features were chosen (Fig. 4 A-B; Table 2 and Supplementary Fig. 1–2). The model showed good performance in evaluating the Ki67 index level, with an AUC of 0.634 for training set, 0.666 for validation set 1, and 0.602 for validation set 2 (Fig. 4 C-E). In addition to AUC values, classification metrics including accuracy, precision, recall, specificity, and F1 score were also evaluated to further assess model performance. In the training set, the model achieved an accuracy of 0.578, a precision of 0.406, a recall of 0.578, a specificity of 0.578, and an F1 score of 0.477. In validation set 1, the corresponding values were: accuracy = 0.579, precision = 0.516, recall = 0.640, specificity = 0.531, and F1 score = 0.571. In the external validation set (validation set 2), the model achieved an accuracy of 0.595, a precision of 0.421, a recall of 0.667, a specificity of 0.560, and an F1 score of 0.516. These results indicate that the model maintained stable and acceptable classification performance across internal and external datasets, reflecting its generalizability and clinical applicability. Figure 4 F-H show the DCA results of the training set and the two validation sets. This analysis revealed that within the threshold probability range of approximately 0–50%, the net benefit derived from using the radiomics model to guide treatment decisions was superior to that of unconditional treatment (the “ALL” strategy) or foregoing treatment entirely (the “None” strategy). This indicates that within this probability range, the model-guided treatment decisions confer substantial health benefits.

We further calculated the Radiomics (Rad) score for each patient and incorporated this score with other pre-operative non-invasive features to generate a Norman plot (Fig. 5 A) and a calibration curve (Fig. 5 B). The results demonstrated that the Rad score outperforms other non-invasive measures in predicting patient outcomes. Furthermore, the calibration curve illustrates the model’s conformity with ideal predictions following bias correction, indicating that the model has great predictive accuracy.

High Ki67 expression predicts better prognosis in patients with gastric cancer

The conventional wisdom is that the higher the Ki67, the worse the patient’s prognosis. However, by grouping 213 patients according to the Ki67 median and plotting a KM curve, we found that the higher the Ki67 index, the longer the survival of patients (Fig. 5 C). To further verify this phenomenon, univariate and multivariate Cox regression analyses were performed to identify variables strongly associated with the prognosis of gastric cancer. The results showed that the Ki67 index was associated with patient prognosis in both univariate and multivariate Cox regression analyses, in univariate analysis (Figure 5 D), Ki67 expression ratio: P = 0.002, HR = 2.370, 95%CI: 1.360–4.130; multivariate analysis (Fig. 5 E), Ki67 expression ratio: P = 0.007, HR = 2.370, 95%CI: 1.270–4.420.

We also calculated the rad score and divided patients into two groups based on the median score and investigated the correlation between the Rad score and Ki67 expression levels (Fig. 5 F). The analysis revealed that patients with higher Ki67 expression generally exhibited higher Rad scores. Furthermore, Kaplan-Meier survival analysis (Fig. 5 G) showed a significant difference in overall survival (OS) between these groups, as determined by the log-rank test ( P = 0.019). During the first 25 months, survival probabilities were similar between the groups; however, as time progressed, survival significantly deteriorated in the low Rad score group (black curve, Group L), whereas it remained relatively stable in the high Rad score group (red curve, Group H). These results demonstrate the radiomics model’s potential utility in forecasting surgical patient outcomes, suggesting that higher rad scores are associated with better patient outcomes, which supports its use in clinical decision making.

Discussion

Ki67 is a well-recognized marker of cellular proliferation and has been widely used in prognostic assessment across various cancer types. In breast cancer, it helps distinguish molecular subtypes and is associated with relapse-free survival and disease-specific survival [ 15 , 16 ]. In prostate cancer, Ki67 combined with prostate-specific antigen (PSA) improves prognostic accuracy [ 17 ], while in multiple myeloma, the Ki67 index in bone marrow serves as a critical predictor of prognosis [ 18 ]. Additionally, its expression has demonstrated prognostic significance in lung cancer, bladder cancer, osteosarcoma, and other malignancies [ 19 – 22 ]. In gastric cancer, Ki67, along with other biomarkers such as β-catenin, Her-2, and PTEN, contributes to a more comprehensive understanding of tumor behavior and treatment response [ 23 ]. These findings underscore the clinical value of Ki67 as a biomarker for patient stratification, treatment selection, and outcome prediction.

Compared with traditional Ki67 biomarker detection methods, radiomics uses non-invasive imaging methods such as MRI and CT to predict Ki67 expression levels, significantly reducing the discomfort and potential complications associated with biopsy. In addition, radiomics has continuous monitoring capabilities that allow clinicians to track tumor response to treatment and possible genetic mutations or behavioral changes in real time, thus avoiding repeat biopsies [ 24 – 27 ]. In our study, we developed a radiomics model based on contrast-enhanced CT images to predict the Ki67 index and prognosis in patients with gastric tubular adenocarcinoma. The model achieved moderate predictive performance, with AUCs of 0.634, 0.666, and 0.602 in the training set, validation set 1, and validation set 2, respectively. It is worth noting that AUC values in the range of 0.60–0.70 are commonly reported in radiomics-based models and are widely accepted as clinically meaningful, particularly in complex tasks such as predicting molecular expression, biomarker levels, or treatment response. Several studies have reported similar AUC values and confirmed their potential for clinical use. For example, the external validation AUC of a model predicting HER2 status in gastric cancer based on CT images was about 0.71 [ 28 ]. In multicenter studies, models using DSC-MRI radiomics to predict glioma IDH mutations had AUC of 0.639 to 0.667 [ 29 ]. A systematic review of predictive models for laryngeal and hypopharyngeal cancer showed an externally validated AUC of 0.67 to 0.73, which is still considered superior to traditional TNM staging systems [ 30 ]. In addition, studies on prediction of acute oral mucous inflammation of nasopharyngeal carcinoma, evaluation of neoadjuvant therapy and survival prediction model of non-small cell lung cancer, prediction of lung function, prediction of glioblastoma molecular subtypes all reported AUC of radiomic model between 0.60 and 0.65, which also showed clinical applicability [ 31 – 35 ].

Due to the high dimensional characteristics of radiomic data, a reasonable dimensionality reduction strategy is essential to prevent model overfitting and improve generalization ability. This study followed the “10:1 sample-variable ratio” principle commonly used in modeling [ 36 , 37 ], and adopted a two-step feature selection process: ICC analysis (threshold > 0.75) was first used to screen out low-repeatability features, followed by LASSO regression to further compress the feature dimensions. In the process of model construction, we evaluated the system performance of the models with different number of features, and found that when the number of features exceeded a certain range, the models showed unstable performance and overfitting trend in the verification set. On the contrary, too few features will easily lead to information loss and underfitting. In the end, we chose to retain 6 features, taking into account the simplicity and stability of the model, and showed balanced and consistent classification performance in both the training set and the two verification sets, without obvious signs of overfitting, verifying the robustness and extensibility of the model. Of particular concern, the model maintained a high recall rate across datasets, illustrating its potential clinical value in identifying high-risk patients with high Ki67 expression. This capability can be further extended to a variety of application scenarios, including: (1) Preoperative patient risk stratification: predicting patients with high Ki67 expression and assisting in decision-making whether neoadjuvant therapy should be intensified; (2) Individualized treatment options: For patients with low Ki67, especially those with comorbidities or poor tolerance, options with less toxic and side effects can be considered; (3) Optimal allocation of treatment resources: Based on the predicted results, intensive treatment and conventional treatment can be reasonably distinguished to avoid over-treatment; (4) Postoperative follow-up management: adjusted postoperative monitoring frequency and intervention plan according to Ki67 prediction level; (5) Multidisciplinary team decision support: as reference information for preoperative evaluation, to improve the scientific and individual degree of treatment path formulation. In summary, although the model in this study is of medium accuracy, it shows certain clinical translational potential in combination with its non-invasive, good generalization ability and stable recall performance.

In this study, we also observed that higher Ki67 expression was associated with better prognosis, a finding that is somewhat different from conventional wisdom. Ki67 is commonly used as a marker of cell proliferation, and its high expression is often considered to indicate a poor clinical prognosis [ 38 , 39 ]. However, recent studies have shown that the prognostic significance of Ki67 is not absolute, but may be affected by various factors such as tumor type, molecular background, treatment mode and study design. It has been noted that Ki67 expression is not consistently associated with traditional pathological stages (such as pTNM), and its co-expression has not shown significant prognostic value in some studies [ 40 ]. In addition, high expression of Ki67 may be associated with certain favorable gene mutations or biomarker states that influence its prognostic implications. Gerdes et al. have emphasized that although a high Ki67 index generally indicates a more aggressive tumor phenotype, its actual prognostic impact depends on the specific cancer type, patient population, and treatment regimen received [ 41 ]. It is worth noting that the definition of “high Ki67 expression” varies greatly among different tumors. For example, in breast cancer, Ki67 index > 20-30% is usually highly expressed [ 42 , 43 ]; In diffuse large B-cell lymphoma, the threshold is 60-80% [ 44 , 45 ]. In hepatocellular carcinoma, high expression is often defined as > 10-20% [ 46 , 47 ]. These findings suggest that although Ki67 is a useful proliferative marker, the prognostic value of Ki67 for tumors may depend on the specific tumor molecular and histological context as well as study design.

Multiple studies also support the idea that patients with high Ki67 expression may be more sensitive to treatment, resulting in better survival outcomes. For example, in myoinvasive bladder cancer, patients with high Ki67 expression had a better cancer-specific survival rate after chemoradiotherapy combined with adjuvant therapy [ 48 ]; In stage II/III colon cancer, high Ki67 expression is associated with better relapse-free survival, especially in patients receiving 5-fluorouracil based adjuvant chemotherapy [ 49 ]. In breast cancer, high Ki67 expression was significantly associated with pathologic complete response (pCR) rates after taxine and anthracycline neoadjuvant chemotherapy, suggesting that it is more likely to serve as an indicator of treatment sensitivity than a single prognostic marker [ 50 , 51 ]. This trend was also observed in gastric cancer. A study from Bahrain showed lower mortality in patients with high Ki67 expression and higher mortality in patients with low Ki67 expression [ 52 ], consistent with the findings in our study. We speculate that patients with gastric tubular adenocarcinoma with high Ki67 expression may be more sensitive to platinum-containing preoperative or postoperative adjuvant chemotherapy regimes and thus have better treatment response and survival outcomes.

Therefore, we believe that Ki67 is not only a marker of cell proliferation, but also may reflect the sensitivity of a tumor to treatment in a specific tumor biological context. Future studies should further explore the regulatory mechanism of Ki67 expression and its interaction with various therapeutic regimes in order to guide more accurate therapeutic strategies. With a deeper understanding of these mechanisms, physicians can more accurately identify patients who are more sensitive to specific chemotherapy drugs based on Ki67 levels, enabling individualized treatment.

Nonetheless, this study has certain limitations. First, although the radiomics model demonstrated potential in predicting Ki67 expression levels in gastric cancer, its overall predictive accuracy remains moderate, highlighting the need for further optimization. Second, the single-center, retrospective design may limit the generalizability of the findings across different populations and geographic regions. Given that pathological characteristics and treatment responses of gastric cancer may vary substantially by region, validation in broader cohorts is warranted. Third, the selection and extraction of radiomics features depend on specific imaging protocols and software platforms, which may pose challenges for reproducibility and clinical translation. Standardization of imaging procedures and feature extraction workflows is thus essential to ensure consistency and applicability. Furthermore, although this study specifically focused on gastric tubular adenocarcinoma, it is important to recognize that distinct molecular subtypes of gastric cancer (e.g., EBV-positive, MSI, GS, CIN) may influence radiomic patterns and prognostic outcomes. These subtypes exhibit different biological behaviors, imaging phenotypes, and treatment responses [ 53 , 54 ], which could significantly affect the performance of radiomics models. However, due to the retrospective nature of the dataset and the lack of molecular subtype classification in most patients, this aspect could not be assessed in the current analysis. Finally, incomplete follow-up data in the current cohort limited our ability to explore associations between radiomics-predicted Ki67 expression and long-term outcomes such as disease-free survival (DFS), which may reduce the comprehensiveness of prognostic evaluation.

Based on these limitations, future research should focus on several key areas. First, large-scale, prospective, multicenter studies are needed to validate model robustness across diverse populations and clinical settings. Second, integrating clinical variables, molecular subtypes, and multi-modality imaging data (e.g., PET/CT, MRI) may enhance model performance and clinical utility. Third, combining radiomics with genomics approaches could help elucidate the biological underpinnings of imaging features and support the development of subtype-specific models for more refined risk stratification. Lastly, improving follow-up data collection will allow more comprehensive assessment of long-term prognostic outcomes such as DFS and OS, thereby better demonstrating the full clinical potential of radiomics models.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Supplementary Material 2

Supplementary Material 3

Supplementary Material 4

Supplementary Material 5

Acknowledgements

Thanks to Professor Qiang Tong for his guidance in this study.

Author contributions

The experimental design was collaboratively developed by QT, JTL and YX. Pathological data collection and organization were meticulously conducted by JTL, QSW, FFC and ZZ. The construction of the predictive model, data analysis, and creation of illustrative figures were handled by JTL, JFY, and QL. JTL was responsible for drafting the manuscript, while QT undertook the task of revising the article.

Funding

The National Natural Science Foundation of China (No. 81172186), the Natural Science Foundation of Hubei Province (No. 2018CFB504), and the Guidance Foundation of Renmin Hospital of Wuhan University (No. RMYD2018M67) all provided funding for this study.

Data availability

The data used and analyzed during the current study are available from the corresponding authors on reasonable request.

Declarations

Ethics approval and consent to participate

This study strictly adhered to the Declaration of Helsinki, and all experimental procedures were reviewed and approved by the Clinical Research Ethics Committee at Renmin Hospital of Wuhan University (Approval No. WDRY2022-K227). Participants in this study provided informed consent, and written informed consent was signed by them or their guardians.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Jiatong Lu, Yong Xiao and Qiushuang Wang contributed equally to this work.

References

1. Liu C, Lu J, An L. Development and validation of nomograms for predicting the prognosis of early and late recurrence of advanced gastric cancer after radical surgery based on post-recurrence survival. Medicine. 2024;103(22):e38376.

2. Liu C, Tao F, Lu J, Park S, An L. Defining nomograms for predicting prognosis of early and late recurrence in gastric cancer patients after radical gastrectomy. Medicine. 2023;102(42):e35585.

3. Sun X, Kaufman PD. Ki-67: more than a proliferation marker. Chromosoma. 2018;127(2):175–86.

4. Menon SS, Guruvayoorappan C, Sakthivel KM, Rasmi RR. Ki-67 protein as a tumour proliferation marker. Clin Chim Acta. 2019;491:39–45.

5. Xiong DD, Zeng CM, Jiang L, Luo DZ, Chen G. Ki-67/MKI67 as a predictive biomarker for clinical outcome in gastric cancer patients: an updated Meta-analysis and systematic review involving 53 studies and 7078 patients. J Cancer. 2019;10(22):5339–54.

6. Fasching PA, Heusinger K, Haeberle L, Niklos M, Hein A, Bayer CM, et al. Ki67, chemotherapy response, and prognosis in breast cancer patients receiving neoadjuvant treatment. BMC Cancer. 2011;11:486.

7. Hashmi AA, Hashmi KA, Irfan M, Khan SM, Edhi MM, Ali JP, et al. Ki67 index in intrinsic breast cancer subtypes and its association with prognostic parameters. BMC Res Notes. 2019;12(1):605.

8. Brown J, Scardo S, Method M, Schlauch D, Misch A, Picard S, et al. A real-world retrospective study of the use of Ki-67 testing and treatment patterns in patients with HR+, HER2- early breast cancer in the united States. BMC Cancer. 2022;22(1):502.

9. Selmani Z, Molimard C, Overs A, Bazan F, Chaigneau L, Dobi E, et al. Low correlation between Ki67 assessed by qRT-PCR in oncotype Dx score and Ki67 assessed by immunohistochemistry. Sci Rep. 2022;12(1):3617.

10. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, et al. Radiomics: the Bridge between medical imaging and personalized medicine. Nat Reviews Clin Oncol. 2017;14(12):749–62.

11. Xie Y, Wang M, Xia H, Sun H, Yuan Y, Jia J, et al. Development and validation of a CECT-based radiomics model for predicting IL1B expression and prognosis of head and neck squamous cell carcinoma. Front Oncol. 2023;13:1121485.

12. Zheng S-J, Zheng C-P, Zhai T-T, Xu X-E, Zheng Y-Q, Li Z-M, et al. Development and validation of a new staging system for esophageal squamous cell carcinoma patients based on combined pathological TNM, radiomics, and proteomics. Ann Surg Oncol. 2022;30(4):2227–41.

13. Jiang Y, Zhou K, Sun Z, Wang H, Xie J, Zhang T et al. Non-invasive tumor microenvironment evaluation and treatment response prediction in gastric cancer using deep learning radiomics. Cell Rep Med. 2023;4(8).

14. Cui Y, Zhang J, Li Z, Wei K, Lei Y, Ren J, et al. A CT-based deep learning radiomics nomogram for predicting the response to neoadjuvant chemotherapy in patients with locally advanced gastric cancer: A multicenter cohort study. EClinicalMedicine. 2022;46:101348.

15. Cheang MC, Chia SK, Voduc D, Gao D, Leung S, Snider J, et al. Ki67 index, HER2 status, and prognosis of patients with luminal B breast cancer. J Natl Cancer Inst. 2009;101(10):736–50.

16. Baskota SU, Dabbs DJ, Clark BZ, Bhargava R. Prosigna ® breast cancer assay: histopathologic correlation, development, and assessment of size, nodal status, Ki-67 (SiNK™) index for breast cancer prognosis. Mod Pathology: Official J United States Can Acad Pathol Inc. 2021;34(1):70–6.

17. Hammarsten P, Josefsson A, Thysell E, Lundholm M, Hägglöf C, Iglesias-Gato D, et al. Immunoreactivity for prostate specific antigen and Ki67 differentiates subgroups of prostate cancer related to outcome. Mod Pathology: Official J United States Can Acad Pathol Inc. 2019;32(9):1310–9.

18. Atrash S, Robinson M, Taneja A, Paul B, Cassetta K, Ndiaye A, et al. Bone marrow Ki-67 index is of prognostic value in newly diagnosed multiple myeloma. Eur J Haematol. 2023;111(3):373–81.

19. Zeng M, Zhou J, Wen L, Zhu Y, Luo Y, Wang W. The relationship between the expression of Ki-67 and the prognosis of osteosarcoma. BMC Cancer. 2021;21(1):210.

20. Qiang Z, Zhang W, Jin S, Dai K, He Y, Tao L, et al. Carcinoembryonic antigen, α-fetoprotein, and Ki67 as biomarkers and prognostic factors in intrahepatic cholangiocarcinoma: A retrospective cohort study. Ann Hepatol. 2021;20:100242.

21. Centonze G, Maisonneuve P, Simbolo M, Lagano V, Grillo F, Fabbri A, et al. Lung carcinoid tumours: histology and Ki-67, the eternal rivalry. Histopathology. 2023;82(2):324–39.

22. Geelvink M, Babmorad A, Maurer A, Stöhr R, Grimm T, Bach C et al. Diagnostic and prognostic implications of FGFR3(high)/Ki67(high) papillary bladder cancers. Int J Mol Sci. 2018;19(9).

23. Wu HW, Qin CY, Huang JL, Kong XY, Wang WJ, Bai WK. Correlations of β-catenin, Ki67 and Her-2/neu with gastric cancer. Asian Pac J Trop Med. 2014;7(4):257–61.

24. Luo X, Zheng R, Zhang J, He J, Luo W, Jiang Z, et al. CT-based radiomics for predicting Ki-67 expression in lung cancer: a systematic review and meta-analysis. Front Oncol. 2024;14:1329801.

25. Deng S, Ding J, Wang H, Mao G, Sun J, Hu J, et al. Deep learning-based radiomic nomograms for predicting Ki67 expression in prostate cancer. BMC Cancer. 2023;23(1):638.

26. Qiao X, Gu X, Liu Y, Shu X, Ai G, Qian S et al. MRI Radiomics-Based machine learning models for Ki67 expression and Gleason grade group prediction in prostate cancer. Cancers. 2023;15(18).

27. Treanor LM, Frank RA, Atyani A, Sharifabadi AD, Hallgrimson Z, Fabiano N, et al. Reporting bias in imaging diagnostic test accuracy studies: are studies with positive conclusions or titles submitted and published faster?? AJR Am J Roentgenol. 2021;216(1):225–32.

28. Shen L, Li Y, Huang H, Lu Z, Chen B. HER2 in gastric cancer: A comprehensive analysis combining Meta-Analysis and DCE-MRI radiomics. Cancer Control: J Moffitt Cancer Cent. 2024;31:10732748241293699.

29. Manikis GC, Ioannidis GS, Siakallis L, Nikiforaki K, Iv M, Vozlic D, et al. Multicenter DSC-MRI-Based radiomics predict IDH mutation in gliomas. Cancers. 2021;13:16.

30. Abou-Foul AK, Dretzke J, Albon E, Kristunas C, Moore DJ, Karwath A, et al. Clinical predictive models for recurrence and survival in treated laryngeal and hypopharyngeal cancer: a systematic review and meta-analysis. Front Oncol. 2024;14:1478385.

31. Ieko Y, Kadoya N, Tanaka S, Kikuchi K, Yamamoto T, Ariga H, et al. Radiomics and dosiomics approaches to estimate lung function after stereotactic body radiation therapy in patients with lung tumors. Radiol Phys Technol. 2025;18(1):238–48.

32. Grossmann P, Gutman DA, Dunn WD Jr., Holder CA, Aerts HJ. Imaging-genomics reveals driving pathways of MRI derived volumetric tumor phenotype features in glioblastoma. BMC Cancer. 2016;16:611.

33. Wang H, Hong Y, Zhang Z, Cheng K, Chen B, Zhang R. Study on postoperative survival prediction model for non-small cell lung cancer: application of radiomics technology workflow based on multi-organ imaging features and various machine learning algorithms. Front Med. 2025;12:1517765.

34. Dong Y, Zhang J, Lam S, Zhang X, Liu A, Teng X et al. Multimodal data integration to predict severe acute oral mucositis of nasopharyngeal carcinoma patients following radiation therapy. Cancers. 2023;15(7).

35. Coroller TP, Agrawal V, Narayan V, Hou Y, Grossmann P, Lee SW, et al. Radiomic phenotype features predict pathological response in non-small cell lung cancer. Radiotherapy Oncology: J Eur Soc Therapeutic Radiol Oncol. 2016;119(3):480–6.

36. Vittinghoff E, McCulloch CE. Relaxing the rule of ten events per variable in logistic and Cox regression. Am J Epidemiol. 2007;165(6):710–8.

37. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol. 1996;49(12):1373–9.

38. Luo G, Hu Y, Zhang Z, Wang P, Luo Z, Lin J, et al. Clinicopathologic significance and prognostic value of Ki-67 expression in patients with gastric cancer: a meta-analysis. Oncotarget. 2017;8(30):50273–83.

39. Yang C, Zhang J, Ding M, Xu K, Li L, Mao L, et al. Ki67 targeted strategies for cancer therapy. Clin Translational Oncology: Official Publication Federation Span Oncol Soc Natl Cancer Inst Mexico. 2018;20(5):570–5.

40. Wei Z, Huang L, Zhang X, Xu A. Expression and significance of Her2 and Ki-67 in gastric adenocarcinoma without distant metastasis: a cohort study. BMC Gastroenterol. 2020;20(1):343.

41. Gerdes J, Lemke H, Baisch H, Wacker HH, Schwab U, Stein H. Cell cycle analysis of a cell proliferation-associated human nuclear antigen defined by the monoclonal antibody Ki-67. Journal of immunology (Baltimore, Md: 1950). 1984;133(4):1710-5.

42. Inwald EC, Klinkhammer-Schalke M, Hofstädter F, Zeman F, Koller M, Gerstenhauer M, et al. Ki-67 is a prognostic parameter in breast cancer patients: results of a large population-based cohort of a cancer registry. Breast Cancer Res Treat. 2013;139(2):539–52.

43. Dowsett M, Nielsen TO, A’Hern R, Bartlett J, Coombes RC, Cuzick J, et al. Assessment of Ki67 in breast cancer: recommendations from the international Ki67 in breast cancer working group. J Natl Cancer Inst. 2011;103(22):1656–64.

44. Xu-Monette ZY, Wu L, Visco C, Tai YC, Tzankov A, Liu WM, et al. Mutational profile and prognostic significance of TP53 in diffuse large B-cell lymphoma patients treated with R-CHOP: report from an international DLBCL Rituximab-CHOP consortium program study. Blood. 2012;120(19):3986–96.

45. Klapper W, Kreuz M, Kohler CW, Burkhardt B, Szczepanowski M, Salaverria I, et al. Patient age at diagnosis is associated with the molecular characteristics of diffuse large B-cell lymphoma. Blood. 2012;119(8):1882–7.

46. Luo JH, Ren B, Keryanov S, Tseng GC, Rao UN, Monga SP, et al. Transcriptomic and genomic analysis of human hepatocellular carcinomas and hepatoblastomas. Hepatology (Baltimore MD). 2006;44(4):1012–24.

47. Calvisi DF, Ladu S, Gorden A, Farina M, Conner EA, Lee JS, et al. Ubiquitous activation of Ras and Jak/Stat pathways in human HCC. Gastroenterology. 2006;130(4):1117–28.

48. Tanabe K, Yoshida S, Koga F, Inoue M, Kobayashi S, Ishioka J, et al. High Ki-67 expression predicts favorable survival in Muscle-Invasive bladder cancer patients treated with Chemoradiation-Based bladder-Sparing protocol. Clin Genitourin Cancer. 2015;13(4):e243–51.

49. Fluge Ø, Gravdal K, Carlsen E, Vonen B, Kjellevold K, Refsum S, et al. Expression of EZH2 and Ki-67 in colorectal cancer and associations with treatment response and prognosis. Br J Cancer. 2009;101(8):1282–9.

50. Wei XX, Zhang R, Pu TJ, Yang LB, Ye F, Bu H. [Ki-67 expression and its effect on response to neo-adjuvant chemotherapy in invasive breast cancer]. Zhonghua Bing Li Xue Za zhi = chinese. J Pathol. 2017;46(12):817–21.

51. Yuan P, Xu B, Wang C, Zhang C, Sun M, Yuan L. Ki-67 expression in luminal type breast cancer and its association with the clinicopathology of the cancer. Oncol Lett. 2016;11(3):2101–5.

52. Awadh M, Darwish A, Alqatari H, Buzaid FM, Darwish A. A descriptive analysis of gastric cancer with an immunohistochemical study of Ki67 and p53 as prognostic factors.: Bahrain experience. Saudi Med J. 2023;44(12):1300–9.

53. Chia NY, Tan P. Molecular classification of gastric cancer. Annals Oncology: Official J Eur Soc Med Oncol. 2016;27(5):763–9.

54. Cristescu R, Lee J, Nebozhyn M, Kim KM, Ting JC, Wong SS, et al. Molecular analysis of gastric cancer identifies subtypes associated with distinct clinical outcomes. Nat Med. 2015;21(5):449–56.

Associated Data

Supplementary Materials

Supplementary Material 1

Supplementary Material 2

Supplementary Material 3

Supplementary Material 4

Supplementary Material 5

Data Availability Statement

The data used and analyzed during the current study are available from the corresponding authors on reasonable request.