Machine learning models in predicting high malignant potential in gastrointestinal stromal tumors: a systematic review and meta-analysis

Machine learning models in predicting high malignant potential in gastrointestinal stromal tumors: a systematic review and meta-analysis | 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 Machine learning models in predicting high malignant potential in gastrointestinal stromal tumors: a systematic review and meta-analysis Xiaojing Wang, Xueyan Guo, Baima Yangji, Lin Shen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5382250/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Gastrointestinal stromal tumors (GIST) present diverse biological behaviors and varying malignancy risks, necessitating early detection and effective risk stratification. Machine learning (ML) models are emerging as valuable tools for assessing the malignant potential of GIST. This systematic review and meta-analysis evaluate the efficacy of various ML models in predicting GIST malignancy. We conducted a comprehensive literature search in PubMed, EMBASE, Web of Science, and the Cochrane Library, adhering to PRISMA guidelines, up to April 22, 2024. After article selection, we extracted essential data and performed meta-analysis to aggregate the c-index, sensitivity, and specificity. The risk of bias was assessed using the PROBAST framework. Our analysis included 12 studies involving 20 ML models and 2,859 patients. Tumor size emerged as the most significant predictor. The pooled c-index was 0.89 (training) and 0.87 (test), with sensitivities of 0.85 and 0.82, and specificities of 0.89 and 0.75, respectively. Two studies had high bias risk, while ten had low bias, although overall applicability was considered low due to inadequate data sources. ML models demonstrate strong diagnostic capabilities in predicting GIST malignancy, with the Logistic Regression model performing best. Key predictive factors included tumor size, necrosis, ulceration, and shape regularity. Future models should integrate impactful disease predictors to enhance clinical utility. Health sciences/Gastroenterology/Gastrointestinal diseases Biological sciences/Computational biology and bioinformatics Gastrointestinal stromal tumors High malignant potential Risk prediction model Systematic review Meta-analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction Gastrointestinal stromal tumors (GIST) are the most common mesenchymal tumors of the gastrointestinal tract, accounting for approximately 1–3% of all gastrointestinal malignancies 1 , 2 . They originate from the interstitial cells of Cajal or related stem cells and can occur anywhere along the gastrointestinal tract, with the stomach (60–70%) and small intestine (20–30%) being the most frequent sites 3 . Although surgical resection is the treatment of choice for GIST, postoperative recurrence and metastasis remain major clinical challenges 4 , 5 , 6 . Predicting the malignant potential of GIST is crucial for developing individualized treatment strategies that can effectively improve patient survival. Traditionally, the prediction of the malignant potential of GIST is mainly based on clinicopathological features, such as tumor size, mitotic index (MI), tumor location and other clinicopathological features 7 , 8 , 9 . However, these indicators have certain limitations and it is difficult to comprehensively and accurately assess the biological behavior of GIST. With the development of medical imaging technology and molecular biology, more and more studies are devoted to predicting the prognosis of GIST through high-throughput data such as radiomics and genomics 10 , 11 , 12 , 13 , 14 . Especially, important results have been achieved in imaging genomics 15 . However, due to the heterogeneity of data, the limitation of sample size, and the complexity of statistical methods, these prediction methods still face many challenges in practical application. Currently available studies focus on GIST recurrence, metastasis, and survival, and there are no studies with a quantitative synthesis of model performance, and there is incomplete understanding of whether the use of ML prediction models in GIST populations provides clinical benefit. Therefore, we conducted a systematic review and meta-analysis to provide a quantitative synthesis of applicable GIST hyper malignant risk prediction models with the aim of improving the accuracy of risk stratification and guiding clinical decision-making. We synthesized the discriminatory power of the included risk prediction models to determine which models may be suitable for clinical use. However, despite the large number of studies that have explored the application of machine learning in the prediction of GIST malignant potential, there is still a lack of systematic reviews and meta-analyses to assess the overall effectiveness and clinical application value of these studies 16 , 17 , 18 , 19 , 20 . Therefore,the aim of this study was to assess and summarize the current status and effectiveness of existing machine learning models in predicting GIST risk factors through meta-analysis, to explore their feasibility and future development direction in clinical practice, and to identify the most commonly used predictors and features in these models. Methods Literature Search Strategies We searched the PubMed, EMBASE, Web of Science and the Cochrane Library from inception to 22 April 2024. We used a combination of medical subject heading terms and free text terms related to "gastrointestinal stromal tumors", "gastric mesenchymal tumor" and "high malignant potential", et al, and the search was limited to the English language. Inclusion/exclusion criteria Articles that meet the following criteria were included: (1) The diagnosis of GIST was confirmed through histopathologic examination; (2) The studies developed or validated a predication model for diagnosis of GIST based on machine learning; (3) The article reported c-index as endpoint or with enough data to infer the c-index; (4) The article clearly described the predictors of high malignant potential; (5) The studies were written in English. Studies meeting the following criteria were excluded: (1) Repeated publications or translations; (2) Articles with incomplete or unavailable data; (3) Letters, animal studies, reviews, conference summaries, and case reports; (4) No detailed description of the modeling process or approach. Study selection and screening The screening process was carried out independently by two authors (A and B). First, duplicate studies were removed, and then the remaining studies were assessed based on the title and abstract content. Next, the full text of the remaining articles was reviewed strictly in accordance with the inclusion and exclusion criteria, and the reference lists of all eligible studies were checked to ensure all the studies were included. Disagreements were resolved by consultation with the third author (C). This meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 21 . This review was registered on PROSPERO (registration no. CRD42024539452). Data Extractions Data extraction was performed independently by two authors (A B). The extraction was designed based on a modified version of the Critical Appraisal and Data Extraction Checklist for the Evaluation of Research Systems for Predictive Modeling (CHARMS) 22 . When discrepancies occurred, they were resolved by the third author (C). The extracted data are as follows. (1) Basic information: first author, year of publication, study design, country, data source, and sample size; (2) Model information: model variables, selection method, model establishment method, model validation method, missing data processing method, predictors finally used in the model; (3) Predicted outcomes: c-index, accuracy, sensitivity and specificity. Quality assessment Assessment of risk of bias and suitability for inclusion in the study was carried out using the Predictive Modeling Risk of Bias Assessment Tool (PROBAST) The PROBAST checklist includes four domains; participants, predictors, outcomes, and analyses 23 . Each study was categorized as "high risk", "low risk" or "unclear risk". Questions in each area could be answered as "yes," "probably yes," "no," "probably not," or "no information". If at least one question in a domain is answered "no" or "may or may not", that domain should be considered at high risk of bias. Only if all domains are judged to be at low risk of bias can the overall bias be considered low risk. Data synthesis and statistical analysis Stata software (version 18.0; Stata Corporation, College Station, Texas, USA) was used for statistical analysis. Subgroups were divided according to ML algorithms. The c-index, accuracy, sensitivity and specificity for prediction in GIST patients, were measured with 95% confidence intervals (95%CIs) in the final analysis. Heterogeneity was tested using the I2 index, which measures heterogeneity with values of 25%, 50%, and 75% indicating low, medium, and high heterogeneity, respectively. Depending on the heterogeneity of the obtained analyses, publication bias was determined using Deeks' Funnel Plot Asymmetry Test, with a fixed-effects or random-effects model when P > 0.05 indicated a low likelihood of publication bias. Results Study selection A total of 4,451 articles were identified through searches in the four databases. After removing 1,319 duplicates and excluding 2,935 articles based on title and abstract review, 197 full-text articles were assessed. Of these, 185 articles were excluded for various reasons (as detailed in Fig. 1 ). Ultimately, 12 studies were included in this review. Characteristics of the included studies The included studies were published between 2019 and 2024, all of which were retrospective studies conducted in China. Among these, three were multi-center studies and nine were single-center studies. The sample sizes ranged from 101 to 494 participants. The machine learning models in included studies A total of 20 models (ranging from 1 to 4 models per study) were retrieved from the included studies. Among them, all studies used logistic regression analysis to build predictive models, except for Jia who used Quadratic Discriminant Analysis (QDA) to build predictive models. Additionally, Support Vector Machine (SVM) and Gradient Boosted Decision Tree (GBDT) were employed in Chen et al.'s study (see Table 1 ). Table 1 Overview of basic data of the included studies. SVM, support vector machine; LR, logistic regression; XGboots, extreme gradient boosting; QDA, quadratic discriminant analysis; GBDT, gradient boosting decision tree. Study County Study design No.patients in the train set No.patients in the test set Technique used for feature selection Type of machine learning Data source Chen, T.(2019) China Retro 130 92 Relief SVM,LR,LR,LR Multiple institution Caiyue Ren(2019) China Retro 117 51 LASSO LR,LR Single institution Chao Wang(2019) China Retro 233 100 stepwise backward LR Single institution Ren, C.(2020) China Retro 308 132 LASSO LR Single institution Li, C.(2021) China Retro 167 39 forward likelihood ratio (LR) LR Single institution Weiqun Ao(2021) China Retro 165 71 LASSO LR,LR Single institution Hairui Chu(2021) China Retro 205 87 LASSO LR Single institution Sun, X. F.(2022) China Retro 69 34 stepwise backward LR, XGboost, linear regression Single institution Liu, L.(2023) China Retro 179 79 LASSO LR Single institution Xiaoxuan Jia(2023) China Retro 74 37 LASSO QDA Multiple institution Cui Zhang(2023) China Retro 161 148 stepwise regression LR,GDBT Multiple institution Liu, Z.(2024) China Retro 345 149 stepwise backward LR Single institution The results of feature selection Feature selection is vital in training machine learning models for predicting the malignant potential of gastrointestinal stromal tumors (GISTs). The number of features used in these models varies from 1 to 10, with 34 commonly used features identified, as shown in Fig. 2 . Key features include tumor size, necrosis, and ulceration, which are strongly associated with malignancy risk. Commonly used predictors in machine learning models In our systematic review and meta-analysis, we identified several predictors that are frequently used in machine learning models to predict the likelihood of GIST malignancy. Tumor size was the most commonly used predictor, with 10 out of 12 studies reporting tumor size. Tumor necrosis was reported in 6 of 12 studies. 5 studies assessed the regularity of tumor shape in the presence of tumor ulceration. Irregular tumor shape and the presence of ulcers with ulceration were associated with more aggressive disease and poorer prognosis. Therefore, it is a valuable predictor in the ML model. Tumor classification has been reported in 4 studies, as shown in Fig. 3 . Risk prediction model performance among included studies The number of training and test models varied, with 11 studies reporting the training and test sets in full, with the training set including 1988 patients and the test set including 871 patients. In the training group, total c-index were 0.89 [0.86, 0.92], I2 = 82.3%. The c-index was 0.89 [95% CI (0.86, 0.92)] for the LR model, 0.98 [95% CI (0.96, 1.00)] for the GBDT model, 0.92 [95% CI (0.83, 1.00)] for the XGboost model, and 0.92 [95% CI (0.82, 1.02)] for the linear regression model, the c-index was 0.87 [95% CI (0.78, 0.9)] for the QDA model, the c-index for the SVM model was 0.87 [95% CI (0.80, 0.93)], and the c-index for the DT model was 0.88 [95% CI (0.83, 0.94)]. I2 = 80.8% for the LR model and I2 = 0.0% for all the rest of the above models. Total I2 = 82.3%, indicating a high degree of heterogeneity.(Fig. 4 ) In the test group, total c-index were 0.87 [0.83, 0.90], I2 = 77.8%. The c-index of the LR model was 0.87 [95% CI (0.83, 0.92)], the c-index of the GBDT model was 0.81 [95% CI (0.71, 0. 92)], the c-index of the XGboost model was 0.88 [95% CI (0.77, 0.99)], and the c-index of the linear regression model was 0.89 [95% CI (0.78, 1.01)], the c-index for the QDA model was 0.81 [95% CI (0.68, 0.92)], the c-index for the SVM model was 0.85 [95% CI (0.76, 0.93)], and the c-index for the DT model was 0.80 [95% CI (0.67, 0.93)]. I2 = 84.0% for the LR model and I2 = 0.0% for all the rest of the above models. (Fig. 5 ) The overall pooled sensitivity for the training group was 0.85 [95% CI (0.79–0.89)], I2 = 70.05%. The overall specificity was 0.82 [95% CI (0.75–0.87)], I2 = 89.38%. The overall pooled sensitivity of the test group was 0.89 [95% CI (0.85–0.92)], I2 = 4.78%. The overall specificity was 0.75 [95% CI (0.68–0.81)], I2 = 82.38%.(Fig. 6 – 7 ) Risk of bias and applicability assessment In conclusion, in terms of overall risk of bias, only 2 studies out of all were rated as high risk of bias because missing data were not handled appropriately. In terms of overall applicability, all studies were rated as low risk of applicability (Table 2 ). The Deeks' funnel plot asymmetry test was symmetrical with the p value > 0.05 which indicates the study has low publication bias.(Fig. 8 – 9 ) Table 2 PROBAST, prediction model risk of bias assessment tool. Author Year Risk of bias Overall risk of bias rating Overall applicability rating Participants Predictors Outcome Analysis Chen, T. 2019 high low low low high low Ren, C. 2020 high low low low high low Li, C. 2021 high low low high high low Sun, X. F. 2022 high low low high high low Liu, L. 2023 high low unclear high high low Liu, Z. 2024 high unclear high high high low Xiaoxuan Jia 2023 high unclear unclear high high low Caiyue Ren 2019 high low low high high low Cui Zhang 2023 high low low high high low Weiqun Ao 2021 high low low low high low Hairui Chu 2021 high low low low high low Chao Wang 2019 high low low low high low Discussion In recent years there has been a gradual increase in the number of studies in which ML has been applied to predict the malignant potential of tumours, and there is a need to conduct a systematic review of published studies in order to provide guidance for future research. This study is the first comprehensive meta-analysis of ML models that predict the likelihood of high malignancy in GIST. Multiple studies agreed that tumor size, necrosis and ulceration were key factors in predicting malignancy 24 , 25 , 26 . These factors have long been recognized in both national and international GIST guidelines as critical determinants of clinical outcomes.For instance, the National Comprehensive Cancer Network (NCCN) and European Society for Medical Oncology (ESMO) guidelines emphasize the significance of tumor size and mitotic index in assessing the risk of GISTs, directly correlating these factors with the likelihood of recurrence and malignancy 27 . Tumour necrosis and ulceration have also been used frequently and have been shown to be associated with more aggressive tumour behaviour, which is consistent with findings in different populations 28 . These predictors have been consistently found in multiple studies, highlighting their importance in the development of robust machine learning models for predicting the malignant potential of GIST.Moreover, the integration of radiomics into ML models has shown substantial promise in enhancing the predictive accuracy of malignancy risk. Radiomics approaches, which analyze quantitative features extracted from imaging modalities, have been increasingly applied in oncology, particularly in gastrointestinal cancers such as gastric and colorectal cancers.Studies in gastric cancer, for example, have demonstrated that combining radiomics with clinical data, like tumor size and lymphovascular invasion, significantly improves the model's ability to predict lymph node metastasis and overall survival 29 , 30 . A notable observation from our meta-analysis is the variability in feature selection across studies, which contributes to the observed heterogeneity. However, a notable observation from our meta-analysis is the variability in feature selection across studies, contributing to the observed heterogeneity.While our analysis focused on core predictors like tumor size and necrosis, original studies often included a broader range of features, such as tumor location, mitotic rate, and genetic mutations (e.g., KIT and PDGFRA mutations)​. These variables are well-documented in GIST guidelines and are considered crucial in stratifying patients for treatment and surveillance. Future studies could benefit from incorporating these additional factors to improve the generalizability and accuracy of ML model.A subgroup analysis was performed since the difference in the order of magnitude characteristics of machine learning models. In the training group, the logistic regression model had a c-index of 0.89, while the GBDT model had a c-index as high as 0.98. The c-index of the other models, such as XGBoost, linear regression, QDA, SVM, and decision tree, ranged from 0.80 to 0.92. The c-index of the models in the test group decreased slightly but still performed well. Although 2 studies were both assessed as high risk of bias in terms of risk of bias, all of them were assessed as low risk in terms of applicability, which implies that these models have good applicability in practical applications. Radiomics involves extracting a number of features from medical images that enable a more detailed and comprehensive assessment of tumour heterogeneity. This ability allows ML models to capture portions of the data missed by a human observer, thereby improving predictive accuracy. The enhanced performance of radiomics-based ML models has been demonstrated in several studies, which have shown that these models can improve the prediction of the malignant potential of GIST beyond conventional imaging assessments. This systematic review and meta-analysis synthesized data from various studies. The pooled analysis indicates that ML models achieve a high c-index and accuracy in both training and test datasets. Radiomics, which extracts quantitative features from medical images, significantly enhances the predictive capability of ML models. The integration of radiomic features with clinical data, such as tumor size and location, improves the model’s accuracy and provides a comprehensive assessment of tumor behavior.Interestingly, when comparing our findings with meta-analyses of prediction models in other gastrointestinal tumors, such as gastric and colorectal cancers, we observe similar trends. For example, in gastric cancer, ML models that integrate clinical, pathological, and radiomics features have been shown to outperform traditional statistical models in predicting outcomes such as lymph node metastasis and survival​. This suggests that the approach of combining diverse data types, including radiomics, could be universally beneficial across different types of gastrointestinal tumors.Compared with traditional methods for assessing the malignant potential of GIST, including histopathological examination and imaging techniques, have limitations in terms of subjectivity and variability. ML models provide a more objective and consistent approach by analyzing large datasets and identifying patterns that may not be detected by human observers. The use of ML in this context addresses the limitations of traditional methods and provide a more accurate and reliable prediction of malignant potential. The implementation of ML models in clinical practice has the potential to revolutionize the management of GIST. Accurate prediction of malignant potential can guide treatment decisions, including the extent of surgical resection and the need for adjuvant therapy. For instance, patients identified as high-risk through ML models may benefit from more aggressive treatment strategies, while those classified as low-risk can avoid unnecessary interventions.In summary, while our meta-analysis reaffirms the significance of certain key predictors, the variability in feature selection among studies suggests the need for more standardized approaches in future research. Incorporating additional variables, as highlighted in GIST guidelines and observed in other gastrointestinal cancer studies, could lead to more comprehensive and accurate predictive models. This could ultimately enhance the clinical utility of ML models in predicting the malignant potential of GISTs and other related tumors. Limitation We conducted a comprehensive and complete search strategy as well as a thorough analysis methodology, and to the best of our knowledge, the present study is the first meta-analysis that evaluated the ML performance in the assessment of prediction model in the GIST patients.of a diagnostic model in the GIST domain. Despite the promising results, several limitations and challenges need to be addressed. Several limitations must be acknowledged.Firstly, a notable limitation of our study is the high heterogeneity observed among the included studies. This heterogeneity could be attributed to the variation in sample sizes, demographic distributions, and the different sets of features and ML models employed across the studies. Such differences can lead to inconsistencies in model performance and reduce the generalizability of the findings. Nevertheless, this heterogeneity may also highlight the complexity and multifactorial nature of predicting GIST malignancy, suggesting that future research should aim to standardize feature selection and model evaluation processes to minimize variability.Additionally, the retrospective nature of most included studies introduces potential biases, such as selection bias and confounding factors. This can impact the reliability of the reported outcomes, underscoring the need for more prospective studies with standardized methodologies.Despite these limitations, the observed heterogeneity could be seen as a critical finding, as it reflects the challenges and opportunities for improving ML models in this field. Addressing these issues in future studies could lead to the development of more robust, generalizable models, ultimately enhancing the accuracy of GIST malignancy predictions. Conclusion ML models are able to integrate multimodal data, including imaging, gene expression and clinical information, to provide more comprehensive and accurate predictions than traditional image analysis and genomics studies 31 . The ability to integrate data allows models to identify complex patterns that are difficult to capture with traditional methods, thereby greatly improving the accuracy of predictions. In addition, machine learning models are efficient and automated, enabling rapid processing of large amounts of data for non-invasive diagnosis, reducing the subjectivity of traditional methods that rely on the experience of experts. Their good adaptability and scalability enable the models to be continuously optimized as data and technology progress 32 . In conclusion, machine learning models provide a more efficient, non-invasive solution for predicting the high malignant potential of GIST and facilitate the development of personalized medicine. However, future research should focus on developing robust, interpretable, and personalized ML models that can guide clinical decision-making and improve patient outcomes. This systematic review and meta-analysis highlight the current state of ML applications in GIST and provide a foundation for future advancements in this rapidly evolving field. Declarations Ethics approval and consent to participate This article does not contain any studies with human participants or animals performed by any of the authors. Competing interests The authors declare no competing interests Author Contribution Concept and design: WXJ and GXY. Acquisition of data: WXJ, BMYJ, and SL. Writing original draft: WXJ. Review and editing: all authors. All authors have made substantial contributions to this work and have approved the final version of the manuscript. Data Availability The datasets analyzed during this study are available from the corresponding author upon reasonable request. References Li, Y., Xie, F., Xiong, Q., Lei, H. & Feng, P. Machine learning for lymph node metastasis prediction in patients with gastric cancer: A systematic review and meta-analysis. Front. Oncol. 12 , 946038 (2022). Miettinen, M. & Lasota, J. Gastrointestinal stromal tumors: pathology and prognosis at different sites. Semin Diagn. Patho . 23 (2), 70–83 (2006). Yin, X. N. et al. Computed tomography radiogenomics: A potential tool for prediction of molecular subtypes in gastric stromal tumor. World J. Gastrointest. Oncol. 16 (4), 1296–1308 (2024). Keung, E. Z., Fairweather, M. & Raut, C. P. The Role of Surgery in Metastatic Gastrointestinal Stromal Tumors. Curr. Treat. Options Oncol. 17 (2), 8 (2016). Mariani, A. et al. Is laparoscopic approach as treatment of large gastric GIST. acceptable? Langenbecks Arch. Surg. 409 (1), 231 (2024). Liu, L. et al. Effects of surgical management for gastrointestinal stromal tumor patients with liver metastasis on survival outcomes. Front. Oncol. 14 , 1289885 (2024). Miettinen, M., Majidi, M. & Lasota, J. Pathology and diagnostic criteria of gastrointestinal stromal tumors (GISTs): a review. Eur. J. Cancer . 38 (Suppl 5), S39–51 (2022). Corless, C. L. et al. PDGFRA mutations in gastrointestinal stromal tumors: frequency, spectrum and in vitro sensitivity to imatinib. J. Clin. Oncol. 23 (23), 5357–5364 (2005). Esteva, A., Kuprel, B. & Novoa, R. Dermatologist-level classification of skin cancer with deep neural networks. Nature 542 , 115–118 (2017). Yan, W., Yu, H., Xu, C., Zeng, M. & Wang, M. The value of a nomogram model based on CT imaging features in differentiating duodenal gastrointestinal stromal tumors from pancreatic head neuroendocrine tumors. Abdom. Radiol. (2024). Ren, C., Wang, S. & Zhang, S. Development and validation of a nomogram based on CT images and 3D texture analysis for preoperative prediction of the malignant potential in gastrointestinal stromal tumors. Cancer Imaging . 20 (1), 5 (2022). Jovanovic, M. M. et al. Possibility of Using Conventional Computed Tomography Features and Histogram Texture Analysis Parameters as Imaging Biomarkers for Preoperative Prediction of High-Risk Gastrointestinal Stromal Tumors of the Stomach. Cancers (Basel) . 15 (24), 5840 (2023). Danti, G. et al. Relationship between diagnostic imaging features and prognostic outcomes in gastrointestinal stromal tumors (GIST). Acta Biomed. 90 (5-S), 9–19 (2019). Yan, J., Zhao, X., Han, S., Wang, T. & Miao, F. Evaluation of Clinical Plus Imaging Features and Multidetector Computed Tomography Texture Analysis in Preoperative Risk Grade Prediction of Small Bowel Gastrointestinal Stromal Tumors. J. Comput. Assist. Tomogr . 42 (5), 714–720 (2018). Rubin, B. P., Heinrich, M. C. & Corless, C. L. Gastrointestinal stromal tumour. Lancet 369 (9574), 1731–1741 (2007). Rengo, M. et al. Development and Validation of Artificial-Intelligence-Based Radiomics Model Using Computed Tomography Features for Preoperative Risk Stratification of Gastrointestinal Stromal Tumors. J. Pers. Med. 13 (5), 717 (2023). Yang, J. et al. Development of a Malignancy Potential Binary Prediction Model Based on Deep Learning for the Mitotic Count of Local Primary Gastrointestinal Stromal Tumors. Korean J. Radiol. 22 (3), 344–353 (2021). Ning, Z. et al. Pattern Classification for Gastrointestinal Stromal Tumors by Integration of Radiomics and Deep Convolutional Features. IEEE J. Biomed. Health Inf. 23 (3), 1181–1191 (2019). Chen, T. et al. Radiomics nomogram for predicting the malignant potential of gastrointestinal stromal tumours preoperatively. Eur. Radiol. 29 (3), 1074–1082 (2019). Zhang, L. et al. Computed tomography-based radiomics model for discriminating the risk stratification of gastrointestinal stromal tumors. Radiol. Med. 125 (5), 465–473 (2020). Moher, D., Liberati, A., Tetzlaff, J. & Altman, D. G. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. 6 (7), e1000097 (2009). Moons, K. G. et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist. PLoS Med , 11 (10):e1001744 (2014). Wolff, R. F. et al. PROBAST: A tool to assess the risk of bias and applicability of prediction model studies. Ann. Intern. Med. 170 (1), 51–58 (2019). Zhao, L. Y. et al. Prognostic Significance of Tumor Size in 2405 Patients With Gastric Cancer: A Retrospective Cohort Study. Med. (Baltim). 94 (50), e2288 (2015). Guo, P. et al. Prognostic value of tumor size in gastric cancer: an analysis of 2,379 patients. Tumour Biol. 34 (2), 1027–1035 (2013). Deng, J. et al. Tumor size as a recommendable variable for accuracy of the prognostic prediction of gastric cancer: a retrospective analysis of 1,521 patients. Ann. Surg. Oncol. 22 (2), 565–572 (2015). Sung, C. O. et al. Tumor size predicts survival in mucinous gastric carcinoma. J. Surg. Oncol. 106 (6), 757–764 (2012). Xu, C. Y., Shen, J. G., Shen, J. Y., Chen, W. J. & Wang, L. B. Ulcer size as a novel indicator marker is correlated with prognosis of ulcerative gastric cancer. Dig. Surg. 26 (4), 312–316 (2009). Bera, K., Braman, N., Gupta, A., Velcheti, V. & Madabhushi, A. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. Nat. Rev. Clin. Oncol. 19 (2), 132–146 (2022). Jiang, B. et al. Deep learning applications in breast cancer histopathological imaging: diagnosis, treatment, and prognosis. Breast Cancer Res. 26 (1), 137 (2024). Lipkova, J. et al. Artificial intelligence for multimodal data integration in oncology. Cancer Cell. 40 (10), 1095–1110 (2022). Wang, F. A., Li, Y. & Zeng, T. Deep Learning of radiology-genomics integration for computational oncology: A mini review. Comput. Struct. Biotechnol. J. 23 , 2708–2716 (2024). Additional Declarations No competing interests reported. Supplementary Files Supplementaryfile.docx Cite Share Download PDF Status: Posted Version 1 posted 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-5382250","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":385367645,"identity":"80ae4d0b-5387-45a1-b55b-6715aefbdb02","order_by":0,"name":"Xiaojing Wang","email":"","orcid":"","institution":"Shaanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiaojing","middleName":"","lastName":"Wang","suffix":""},{"id":385367648,"identity":"e9802123-a296-441f-9366-225809cedee4","order_by":1,"name":"Xueyan Guo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIiWNgGAWjYBACNv7m4x8+VNjY2R9vPvggoaKGsBY+iWNpjDPOpCUznDmWbPDgzDHCWuQYcsyYedsOMzbc8DGTfNjCTITDGA6YPeY5k8bMOIMtrSKxgY2Bv707Ab8W5oZ0wzkVNnzM0s3HbiTukGGQOHN2AyFbDki8AdrCJnMs7UbiGTYGA4lcQloSGyRAfumRyDErSGxjJkZLMpskSMsMoBYG4rRIHGM2BAWyAc+xZImEM8d4CPpFvr//4wNQVBqwNx/8+KOiRo6/vRe/FgzAQ5ryUTAKRsEoGAVYAQCCN05To7/aqAAAAABJRU5ErkJggg==","orcid":"","institution":"Shaanxi Provincial People's Hospital","correspondingAuthor":true,"prefix":"","firstName":"Xueyan","middleName":"","lastName":"Guo","suffix":""},{"id":385367649,"identity":"5777af5c-8846-4714-bbb3-aa5e0aa6e56f","order_by":2,"name":"Baima Yangji","email":"","orcid":"","institution":"Shaanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Baima","middleName":"","lastName":"Yangji","suffix":""},{"id":385367650,"identity":"e388e2bc-3ded-42d2-99cf-e245c6177ef0","order_by":3,"name":"Lin Shen","email":"","orcid":"","institution":"Shaanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Lin","middleName":"","lastName":"Shen","suffix":""}],"badges":[],"createdAt":"2024-11-03 13:38:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5382250/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5382250/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":71882267,"identity":"4e28e6a0-da34-4ea4-9c9c-b280d7b201d7","added_by":"auto","created_at":"2024-12-19 11:42:44","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":117094,"visible":true,"origin":"","legend":"\u003cp\u003eFlow diagram of study selection\u003c/p\u003e","description":"","filename":"Fig.1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/a777e655efc0936f4004a222.jpg"},{"id":71882266,"identity":"f753a755-c876-4202-85e9-c93a94f11a03","added_by":"auto","created_at":"2024-12-19 11:42:44","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":137600,"visible":true,"origin":"","legend":"\u003cp\u003eRanking of the importance of the features included in the study (importance refers to the sum of the number of times the feature appears in the included articles).\u003c/p\u003e","description":"","filename":"Fig.2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/bdd7ff2ba416bd49d1893a6a.jpg"},{"id":71882238,"identity":"15e75d92-f89a-4489-9034-cd674706a93a","added_by":"auto","created_at":"2024-12-19 11:42:42","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":39905,"visible":true,"origin":"","legend":"\u003cp\u003eMost commonly used predictors in the included studies.\u003c/p\u003e","description":"","filename":"Fig.3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/77421e383099fde1218ccccc.jpg"},{"id":71882274,"identity":"8bc5f012-43c8-4dc2-9863-296491fffbb2","added_by":"auto","created_at":"2024-12-19 11:42:44","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":67437,"visible":true,"origin":"","legend":"\u003cp\u003eC-index of the training group.\u003c/p\u003e","description":"","filename":"Fig.4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/08facfb4a5054c9b2c9f8af0.jpg"},{"id":71882241,"identity":"558edd53-0e75-40d1-b467-3d112a21fc1c","added_by":"auto","created_at":"2024-12-19 11:42:42","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":71504,"visible":true,"origin":"","legend":"\u003cp\u003eC-index of the test group.\u003c/p\u003e","description":"","filename":"Fig.5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/9104b36015ea786b072052e5.jpg"},{"id":71882259,"identity":"2586a098-8c0c-43bb-837b-6858e827f62a","added_by":"auto","created_at":"2024-12-19 11:42:43","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":68626,"visible":true,"origin":"","legend":"\u003cp\u003eSensitivity and specificity for the training group.\u003c/p\u003e","description":"","filename":"Fig.6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/1b0e12a3fdf711e76b4a4868.jpg"},{"id":71882242,"identity":"4b003362-7888-40a4-9e7e-6391c68b9289","added_by":"auto","created_at":"2024-12-19 11:42:42","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":65029,"visible":true,"origin":"","legend":"\u003cp\u003eSensitivity and specificity for the test group.\u003c/p\u003e","description":"","filename":"Fig.7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/20d8d2c275441f995a732152.jpg"},{"id":71882255,"identity":"1f3ffe1e-8e1f-4b7b-8fa0-6dc9d390383f","added_by":"auto","created_at":"2024-12-19 11:42:43","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":36765,"visible":true,"origin":"","legend":"\u003cp\u003eDeeks' funnel plot asymmetry test.\u003c/p\u003e","description":"","filename":"Fig.8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/32e2dffca71071dad57003b3.jpg"},{"id":71882261,"identity":"a8919a4c-0212-444e-8095-f4df175ae5c5","added_by":"auto","created_at":"2024-12-19 11:42:43","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":36714,"visible":true,"origin":"","legend":"\u003cp\u003eDeeks' funnel plot asymmetry test.\u003c/p\u003e","description":"","filename":"Fig.9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/a2905a4a472ffc35a6797def.jpg"},{"id":71884537,"identity":"6a6ffac0-a617-4668-aa80-c8f79673d5f1","added_by":"auto","created_at":"2024-12-19 12:06:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1337681,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/24bfb947-3925-4d22-99d5-3210a4141475.pdf"},{"id":71882237,"identity":"bc00a2fc-4902-46ca-8145-eca7d802b3f8","added_by":"auto","created_at":"2024-12-19 11:42:42","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2666891,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfile.docx","url":"https://assets-eu.researchsquare.com/files/rs-5382250/v1/d1fc44617d0d7ac12f450695.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine learning models in predicting high malignant potential in gastrointestinal stromal tumors: a systematic review and meta-analysis","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGastrointestinal stromal tumors (GIST) are the most common mesenchymal tumors of the gastrointestinal tract, accounting for approximately 1\u0026ndash;3% of all gastrointestinal malignancies\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. They originate from the interstitial cells of Cajal or related stem cells and can occur anywhere along the gastrointestinal tract, with the stomach (60\u0026ndash;70%) and small intestine (20\u0026ndash;30%) being the most frequent sites\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Although surgical resection is the treatment of choice for GIST, postoperative recurrence and metastasis remain major clinical challenges\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e,\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. Predicting the malignant potential of GIST is crucial for developing individualized treatment strategies that can effectively improve patient survival.\u003c/p\u003e \u003cp\u003eTraditionally, the prediction of the malignant potential of GIST is mainly based on clinicopathological features, such as tumor size, mitotic index (MI), tumor location and other clinicopathological features\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. However, these indicators have certain limitations and it is difficult to comprehensively and accurately assess the biological behavior of GIST. With the development of medical imaging technology and molecular biology, more and more studies are devoted to predicting the prognosis of GIST through high-throughput data such as radiomics and genomics\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Especially, important results have been achieved in imaging genomics\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. However, due to the heterogeneity of data, the limitation of sample size, and the complexity of statistical methods, these prediction methods still face many challenges in practical application.\u003c/p\u003e \u003cp\u003eCurrently available studies focus on GIST recurrence, metastasis, and survival, and there are no studies with a quantitative synthesis of model performance, and there is incomplete understanding of whether the use of ML prediction models in GIST populations provides clinical benefit. Therefore, we conducted a systematic review and meta-analysis to provide a quantitative synthesis of applicable GIST hyper malignant risk prediction models with the aim of improving the accuracy of risk stratification and guiding clinical decision-making. We synthesized the discriminatory power of the included risk prediction models to determine which models may be suitable for clinical use.\u003c/p\u003e \u003cp\u003eHowever, despite the large number of studies that have explored the application of machine learning in the prediction of GIST malignant potential, there is still a lack of systematic reviews and meta-analyses to assess the overall effectiveness and clinical application value of these studies\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e,\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Therefore,the aim of this study was to assess and summarize the current status and effectiveness of existing machine learning models in predicting GIST risk factors through meta-analysis, to explore their feasibility and future development direction in clinical practice, and to identify the most commonly used predictors and features in these models.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eLiterature Search Strategies\u003c/h2\u003e \u003cp\u003eWe searched the PubMed, EMBASE, Web of Science and the Cochrane Library from inception to 22 April 2024. We used a combination of medical subject heading terms and free text terms related to \"gastrointestinal stromal tumors\", \"gastric mesenchymal tumor\" and \"high malignant potential\", et al, and the search was limited to the English language.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eInclusion/exclusion criteria\u003c/h3\u003e\n\u003cp\u003eArticles that meet the following criteria were included:\u003c/p\u003e \u003cp\u003e(1) The diagnosis of GIST was confirmed through histopathologic examination;\u003c/p\u003e \u003cp\u003e(2) The studies developed or validated a predication model for diagnosis of GIST based on machine learning;\u003c/p\u003e \u003cp\u003e(3) The article reported c-index as endpoint or with enough data to infer the c-index;\u003c/p\u003e \u003cp\u003e(4) The article clearly described the predictors of high malignant potential;\u003c/p\u003e \u003cp\u003e(5) The studies were written in English.\u003c/p\u003e\n\u003ch3\u003eStudies meeting the following criteria were excluded:\u003c/h3\u003e\n\u003cp\u003e(1) Repeated publications or translations;\u003c/p\u003e \u003cp\u003e(2) Articles with incomplete or unavailable data;\u003c/p\u003e \u003cp\u003e(3) Letters, animal studies, reviews, conference summaries, and case reports;\u003c/p\u003e \u003cp\u003e(4) No detailed description of the modeling process or approach.\u003c/p\u003e\n\u003ch3\u003eStudy selection and screening\u003c/h3\u003e\n\u003cp\u003eThe screening process was carried out independently by two authors (A and B). First, duplicate studies were removed, and then the remaining studies were assessed based on the title and abstract content. Next, the full text of the remaining articles was reviewed strictly in accordance with the inclusion and exclusion criteria, and the reference lists of all eligible studies were checked to ensure all the studies were included. Disagreements were resolved by consultation with the third author (C). This meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. This review was registered on PROSPERO (registration no. CRD42024539452).\u003c/p\u003e\n\u003ch3\u003eData Extractions\u003c/h3\u003e\n\u003cp\u003eData extraction was performed independently by two authors (A B). The extraction was designed based on a modified version of the Critical Appraisal and Data Extraction Checklist for the Evaluation of Research Systems for Predictive Modeling (CHARMS)\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. When discrepancies occurred, they were resolved by the third author (C). The extracted data are as follows.\u003c/p\u003e \u003cp\u003e(1) Basic information: first author, year of publication, study design, country, data source, and sample size;\u003c/p\u003e \u003cp\u003e(2) Model information: model variables, selection method, model establishment method, model validation method, missing data processing method, predictors finally used in the model;\u003c/p\u003e \u003cp\u003e(3) Predicted outcomes: c-index, accuracy, sensitivity and specificity.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eQuality assessment\u003c/h2\u003e \u003cp\u003eAssessment of risk of bias and suitability for inclusion in the study was carried out using the Predictive Modeling Risk of Bias Assessment Tool (PROBAST) The PROBAST checklist includes four domains; participants, predictors, outcomes, and analyses\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. Each study was categorized as \"high risk\", \"low risk\" or \"unclear risk\". Questions in each area could be answered as \"yes,\" \"probably yes,\" \"no,\" \"probably not,\" or \"no information\". If at least one question in a domain is answered \"no\" or \"may or may not\", that domain should be considered at high risk of bias. Only if all domains are judged to be at low risk of bias can the overall bias be considered low risk.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData synthesis and statistical analysis\u003c/h3\u003e\n\u003cp\u003eStata software (version 18.0; Stata Corporation, College Station, Texas, USA) was used for statistical analysis. Subgroups were divided according to ML algorithms. The c-index, accuracy, sensitivity and specificity for prediction in GIST patients, were measured with 95% confidence intervals (95%CIs) in the final analysis. Heterogeneity was tested using the I2 index, which measures heterogeneity with values of 25%, 50%, and 75% indicating low, medium, and high heterogeneity, respectively. Depending on the heterogeneity of the obtained analyses, publication bias was determined using Deeks' Funnel Plot Asymmetry Test, with a fixed-effects or random-effects model when P\u0026thinsp;\u0026gt;\u0026thinsp;0.05 indicated a low likelihood of publication bias.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eStudy selection\u003c/h2\u003e \u003cp\u003eA total of 4,451 articles were identified through searches in the four databases. After removing 1,319 duplicates and excluding 2,935 articles based on title and abstract review, 197 full-text articles were assessed. Of these, 185 articles were excluded for various reasons (as detailed in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Ultimately, 12 studies were included in this review.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eCharacteristics of the included studies\u003c/h2\u003e \u003cp\u003eThe included studies were published between 2019 and 2024, all of which were retrospective studies conducted in China. Among these, three were multi-center studies and nine were single-center studies. The sample sizes ranged from 101 to 494 participants.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eThe machine learning models in included studies\u003c/h2\u003e \u003cp\u003eA total of 20 models (ranging from 1 to 4 models per study) were retrieved from the included studies. Among them, all studies used logistic regression analysis to build predictive models, except for Jia who used Quadratic Discriminant Analysis (QDA) to build predictive models. Additionally, Support Vector Machine (SVM) and Gradient Boosted Decision Tree (GBDT) were employed in Chen et al.'s study (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\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\u003eOverview of basic data of the included studies. SVM, support vector machine; LR, logistic regression; XGboots, extreme gradient boosting; QDA, quadratic discriminant analysis; GBDT, gradient boosting decision tree.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\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=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" 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 \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCounty\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStudy design\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNo.patients in the train set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNo.patients in the test set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTechnique used for feature selection\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eType of machine learning\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eData source\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen, T.(2019)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e130\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eRelief\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSVM,LR,LR,LR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMultiple institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCaiyue Ren(2019)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e117\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR,LR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChao Wang(2019)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e233\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003estepwise backward\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRen, C.(2020)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e308\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e132\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLi, C.(2021)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eforward likelihood ratio (LR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeiqun Ao(2021)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e165\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR,LR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHairui Chu(2021)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e205\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSun, X. F.(2022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003estepwise backward\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR, XGboost, linear regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiu, L.(2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e179\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXiaoxuan Jia(2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eQDA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMultiple institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCui Zhang(2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e161\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003estepwise regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR,GDBT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMultiple institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiu, Z.(2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e345\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e149\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003estepwise backward\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSingle institution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eThe results of feature selection\u003c/h2\u003e \u003cp\u003eFeature selection is vital in training machine learning models for predicting the malignant potential of gastrointestinal stromal tumors (GISTs). The number of features used in these models varies from 1 to 10, with 34 commonly used features identified, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Key features include tumor size, necrosis, and ulceration, which are strongly associated with malignancy risk.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eCommonly used predictors in machine learning models\u003c/h2\u003e \u003cp\u003eIn our systematic review and meta-analysis, we identified several predictors that are frequently used in machine learning models to predict the likelihood of GIST malignancy.\u003c/p\u003e \u003cp\u003eTumor size was the most commonly used predictor, with 10 out of 12 studies reporting tumor size. Tumor necrosis was reported in 6 of 12 studies. 5 studies assessed the regularity of tumor shape in the presence of tumor ulceration. Irregular tumor shape and the presence of ulcers with ulceration were associated with more aggressive disease and poorer prognosis. Therefore, it is a valuable predictor in the ML model. Tumor classification has been reported in 4 studies, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eRisk prediction model performance among included studies\u003c/h2\u003e \u003cp\u003eThe number of training and test models varied, with 11 studies reporting the training and test sets in full, with the training set including 1988 patients and the test set including 871 patients. In the training group, total c-index were 0.89 [0.86, 0.92], I2\u0026thinsp;=\u0026thinsp;82.3%. The c-index was 0.89 [95% CI (0.86, 0.92)] for the LR model, 0.98 [95% CI (0.96, 1.00)] for the GBDT model, 0.92 [95% CI (0.83, 1.00)] for the XGboost model, and 0.92 [95% CI (0.82, 1.02)] for the linear regression model, the c-index was 0.87 [95% CI (0.78, 0.9)] for the QDA model, the c-index for the SVM model was 0.87 [95% CI (0.80, 0.93)], and the c-index for the DT model was 0.88 [95% CI (0.83, 0.94)]. I2\u0026thinsp;=\u0026thinsp;80.8% for the LR model and I2\u0026thinsp;=\u0026thinsp;0.0% for all the rest of the above models. Total I2\u0026thinsp;=\u0026thinsp;82.3%, indicating a high degree of heterogeneity.(Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e )\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn the test group, total c-index were 0.87 [0.83, 0.90], I2\u0026thinsp;=\u0026thinsp;77.8%. The c-index of the LR model was 0.87 [95% CI (0.83, 0.92)], the c-index of the GBDT model was 0.81 [95% CI (0.71, 0. 92)], the c-index of the XGboost model was 0.88 [95% CI (0.77, 0.99)], and the c-index of the linear regression model was 0.89 [95% CI (0.78, 1.01)], the c-index for the QDA model was 0.81 [95% CI (0.68, 0.92)], the c-index for the SVM model was 0.85 [95% CI (0.76, 0.93)], and the c-index for the DT model was 0.80 [95% CI (0.67, 0.93)]. I2\u0026thinsp;=\u0026thinsp;84.0% for the LR model and I2\u0026thinsp;=\u0026thinsp;0.0% for all the rest of the above models. (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e )\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe overall pooled sensitivity for the training group was 0.85 [95% CI (0.79\u0026ndash;0.89)], I2\u0026thinsp;=\u0026thinsp;70.05%. The overall specificity was 0.82 [95% CI (0.75\u0026ndash;0.87)], I2\u0026thinsp;=\u0026thinsp;89.38%. The overall pooled sensitivity of the test group was 0.89 [95% CI (0.85\u0026ndash;0.92)], I2\u0026thinsp;=\u0026thinsp;4.78%. The overall specificity was 0.75 [95% CI (0.68\u0026ndash;0.81)], I2\u0026thinsp;=\u0026thinsp;82.38%.(Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e )\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eRisk of bias and applicability assessment\u003c/h2\u003e \u003cp\u003eIn conclusion, in terms of overall risk of bias, only 2 studies out of all were rated as high risk of bias because missing data were not handled appropriately. In terms of overall applicability, all studies were rated as low risk of applicability (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The Deeks' funnel plot asymmetry test was symmetrical with the p value\u0026thinsp;\u0026gt;\u0026thinsp;0.05 which indicates the study has low publication bias.(Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e )\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\u003ePROBAST, prediction model risk of bias assessment tool.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" 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 \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYear\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c6\" namest=\"c3\"\u003e \u003cp\u003eRisk of bias\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eOverall risk of bias rating\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eOverall applicability rating\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eParticipants\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePredictors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOutcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAnalysis\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen, T.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRen, C.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLi, C.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSun, X. F.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiu, L.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eunclear\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiu, Z.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eunclear\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXiaoxuan Jia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eunclear\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eunclear\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCaiyue Ren\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCui Zhang\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeiqun Ao\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHairui Chu\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChao Wang\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ehigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003elow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn recent years there has been a gradual increase in the number of studies in which ML has been applied to predict the malignant potential of tumours, and there is a need to conduct a systematic review of published studies in order to provide guidance for future research. This study is the first comprehensive meta-analysis of ML models that predict the likelihood of high malignancy in GIST. Multiple studies agreed that tumor size, necrosis and ulceration were key factors in predicting malignancy\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. These factors have long been recognized in both national and international GIST guidelines as critical determinants of clinical outcomes.For instance, the National Comprehensive Cancer Network (NCCN) and European Society for Medical Oncology (ESMO) guidelines emphasize the significance of tumor size and mitotic index in assessing the risk of GISTs, directly correlating these factors with the likelihood of recurrence and malignancy\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Tumour necrosis and ulceration have also been used frequently and have been shown to be associated with more aggressive tumour behaviour, which is consistent with findings in different populations\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. These predictors have been consistently found in multiple studies, highlighting their importance in the development of robust machine learning models for predicting the malignant potential of GIST.Moreover, the integration of radiomics into ML models has shown substantial promise in enhancing the predictive accuracy of malignancy risk. Radiomics approaches, which analyze quantitative features extracted from imaging modalities, have been increasingly applied in oncology, particularly in gastrointestinal cancers such as gastric and colorectal cancers.Studies in gastric cancer, for example, have demonstrated that combining radiomics with clinical data, like tumor size and lymphovascular invasion, significantly improves the model's ability to predict lymph node metastasis and overall survival\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. A notable observation from our meta-analysis is the variability in feature selection across studies, which contributes to the observed heterogeneity. However, a notable observation from our meta-analysis is the variability in feature selection across studies, contributing to the observed heterogeneity.While our analysis focused on core predictors like tumor size and necrosis, original studies often included a broader range of features, such as tumor location, mitotic rate, and genetic mutations (e.g., KIT and PDGFRA mutations)​. These variables are well-documented in GIST guidelines and are considered crucial in stratifying patients for treatment and surveillance. Future studies could benefit from incorporating these additional factors to improve the generalizability and accuracy of ML model.A subgroup analysis was performed since the difference in the order of magnitude characteristics of machine learning models. In the training group, the logistic regression model had a c-index of 0.89, while the GBDT model had a c-index as high as 0.98. The c-index of the other models, such as XGBoost, linear regression, QDA, SVM, and decision tree, ranged from 0.80 to 0.92. The c-index of the models in the test group decreased slightly but still performed well. Although 2 studies were both assessed as high risk of bias in terms of risk of bias, all of them were assessed as low risk in terms of applicability, which implies that these models have good applicability in practical applications. Radiomics involves extracting a number of features from medical images that enable a more detailed and comprehensive assessment of tumour heterogeneity. This ability allows ML models to capture portions of the data missed by a human observer, thereby improving predictive accuracy. The enhanced performance of radiomics-based ML models has been demonstrated in several studies, which have shown that these models can improve the prediction of the malignant potential of GIST beyond conventional imaging assessments.\u003c/p\u003e \u003cp\u003eThis systematic review and meta-analysis synthesized data from various studies. The pooled analysis indicates that ML models achieve a high c-index and accuracy in both training and test datasets. Radiomics, which extracts quantitative features from medical images, significantly enhances the predictive capability of ML models. The integration of radiomic features with clinical data, such as tumor size and location, improves the model\u0026rsquo;s accuracy and provides a comprehensive assessment of tumor behavior.Interestingly, when comparing our findings with meta-analyses of prediction models in other gastrointestinal tumors, such as gastric and colorectal cancers, we observe similar trends. For example, in gastric cancer, ML models that integrate clinical, pathological, and radiomics features have been shown to outperform traditional statistical models in predicting outcomes such as lymph node metastasis and survival​. This suggests that the approach of combining diverse data types, including radiomics, could be universally beneficial across different types of gastrointestinal tumors.Compared with traditional methods for assessing the malignant potential of GIST, including histopathological examination and imaging techniques, have limitations in terms of subjectivity and variability. ML models provide a more objective and consistent approach by analyzing large datasets and identifying patterns that may not be detected by human observers. The use of ML in this context addresses the limitations of traditional methods and provide a more accurate and reliable prediction of malignant potential. The implementation of ML models in clinical practice has the potential to revolutionize the management of GIST. Accurate prediction of malignant potential can guide treatment decisions, including the extent of surgical resection and the need for adjuvant therapy. For instance, patients identified as high-risk through ML models may benefit from more aggressive treatment strategies, while those classified as low-risk can avoid unnecessary interventions.In summary, while our meta-analysis reaffirms the significance of certain key predictors, the variability in feature selection among studies suggests the need for more standardized approaches in future research. Incorporating additional variables, as highlighted in GIST guidelines and observed in other gastrointestinal cancer studies, could lead to more comprehensive and accurate predictive models. This could ultimately enhance the clinical utility of ML models in predicting the malignant potential of GISTs and other related tumors.\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eLimitation\u003c/h2\u003e \u003cp\u003eWe conducted a comprehensive and complete search strategy as well as a thorough analysis methodology, and to the best of our knowledge, the present study is the first meta-analysis that evaluated the ML performance in the assessment of prediction model in the GIST patients.of a diagnostic model in the GIST domain. Despite the promising results, several limitations and challenges need to be addressed. Several limitations must be acknowledged.Firstly, a notable limitation of our study is the high heterogeneity observed among the included studies. This heterogeneity could be attributed to the variation in sample sizes, demographic distributions, and the different sets of features and ML models employed across the studies. Such differences can lead to inconsistencies in model performance and reduce the generalizability of the findings. Nevertheless, this heterogeneity may also highlight the complexity and multifactorial nature of predicting GIST malignancy, suggesting that future research should aim to standardize feature selection and model evaluation processes to minimize variability.Additionally, the retrospective nature of most included studies introduces potential biases, such as selection bias and confounding factors. This can impact the reliability of the reported outcomes, underscoring the need for more prospective studies with standardized methodologies.Despite these limitations, the observed heterogeneity could be seen as a critical finding, as it reflects the challenges and opportunities for improving ML models in this field. Addressing these issues in future studies could lead to the development of more robust, generalizable models, ultimately enhancing the accuracy of GIST malignancy predictions.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eML models are able to integrate multimodal data, including imaging, gene expression and clinical information, to provide more comprehensive and accurate predictions than traditional image analysis and genomics studies\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. The ability to integrate data allows models to identify complex patterns that are difficult to capture with traditional methods, thereby greatly improving the accuracy of predictions. In addition, machine learning models are efficient and automated, enabling rapid processing of large amounts of data for non-invasive diagnosis, reducing the subjectivity of traditional methods that rely on the experience of experts. Their good adaptability and scalability enable the models to be continuously optimized as data and technology progress\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. In conclusion, machine learning models provide a more efficient, non-invasive solution for predicting the high malignant potential of GIST and facilitate the development of personalized medicine. However, future research should focus on developing robust, interpretable, and personalized ML models that can guide clinical decision-making and improve patient outcomes. This systematic review and meta-analysis highlight the current state of ML applications in GIST and provide a foundation for future advancements in this rapidly evolving field.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e \u003cp\u003eThis article does not contain any studies with human participants or animals performed by any of the authors.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eCompeting interests\u003c/strong\u003e \u003cp\u003eThe authors declare no competing interests\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConcept and design: WXJ and GXY. Acquisition of data: WXJ, BMYJ, and SL. Writing original draft: WXJ. Review and editing: all authors. All authors have made substantial contributions to this work and have approved the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets analyzed during this study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eLi, Y., Xie, F., Xiong, Q., Lei, H. \u0026amp; Feng, P. Machine learning for lymph node metastasis prediction in patients with gastric cancer: A systematic review and meta-analysis. \u003cem\u003eFront. Oncol.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 946038 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiettinen, M. \u0026amp; Lasota, J. Gastrointestinal stromal tumors: pathology and prognosis at different sites. \u003cem\u003eSemin Diagn. Patho\u003c/em\u003e. \u003cb\u003e23\u003c/b\u003e (2), 70\u0026ndash;83 (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYin, X. N. et al. Computed tomography radiogenomics: A potential tool for prediction of molecular subtypes in gastric stromal tumor. \u003cem\u003eWorld J. Gastrointest. Oncol.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e (4), 1296\u0026ndash;1308 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeung, E. Z., Fairweather, M. \u0026amp; Raut, C. P. The Role of Surgery in Metastatic Gastrointestinal Stromal Tumors. \u003cem\u003eCurr. Treat. Options Oncol.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e (2), 8 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMariani, A. et al. Is laparoscopic approach as treatment of large gastric GIST. \u003cem\u003eacceptable? Langenbecks Arch. Surg.\u003c/em\u003e \u003cb\u003e409\u003c/b\u003e (1), 231 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, L. et al. Effects of surgical management for gastrointestinal stromal tumor patients with liver metastasis on survival outcomes. \u003cem\u003eFront. Oncol.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 1289885 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiettinen, M., Majidi, M. \u0026amp; Lasota, J. Pathology and diagnostic criteria of gastrointestinal stromal tumors (GISTs): a review. \u003cem\u003eEur. J. Cancer\u003c/em\u003e. \u003cb\u003e38\u003c/b\u003e (Suppl 5), S39\u0026ndash;51 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCorless, C. L. et al. PDGFRA mutations in gastrointestinal stromal tumors: frequency, spectrum and in vitro sensitivity to imatinib. \u003cem\u003eJ. Clin. Oncol.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e (23), 5357\u0026ndash;5364 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEsteva, A., Kuprel, B. \u0026amp; Novoa, R. Dermatologist-level classification of skin cancer with deep neural networks. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e542\u003c/b\u003e, 115\u0026ndash;118 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan, W., Yu, H., Xu, C., Zeng, M. \u0026amp; Wang, M. The value of a nomogram model based on CT imaging features in differentiating duodenal gastrointestinal stromal tumors from pancreatic head neuroendocrine tumors. \u003cem\u003eAbdom. Radiol.\u003c/em\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRen, C., Wang, S. \u0026amp; Zhang, S. Development and validation of a nomogram based on CT images and 3D texture analysis for preoperative prediction of the malignant potential in gastrointestinal stromal tumors. \u003cem\u003eCancer Imaging\u003c/em\u003e. \u003cb\u003e20\u003c/b\u003e (1), 5 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJovanovic, M. M. et al. Possibility of Using Conventional Computed Tomography Features and Histogram Texture Analysis Parameters as Imaging Biomarkers for Preoperative Prediction of High-Risk Gastrointestinal Stromal Tumors of the Stomach. \u003cem\u003eCancers (Basel)\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e (24), 5840 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDanti, G. et al. Relationship between diagnostic imaging features and prognostic outcomes in gastrointestinal stromal tumors (GIST). \u003cem\u003eActa Biomed.\u003c/em\u003e \u003cb\u003e90\u003c/b\u003e (5-S), 9\u0026ndash;19 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan, J., Zhao, X., Han, S., Wang, T. \u0026amp; Miao, F. Evaluation of Clinical Plus Imaging Features and Multidetector Computed Tomography Texture Analysis in Preoperative Risk Grade Prediction of Small Bowel Gastrointestinal Stromal Tumors. \u003cem\u003eJ. Comput. Assist. Tomogr\u003c/em\u003e. \u003cb\u003e42\u003c/b\u003e (5), 714\u0026ndash;720 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRubin, B. P., Heinrich, M. C. \u0026amp; Corless, C. L. Gastrointestinal stromal tumour. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e369\u003c/b\u003e (9574), 1731\u0026ndash;1741 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRengo, M. et al. Development and Validation of Artificial-Intelligence-Based Radiomics Model Using Computed Tomography Features for Preoperative Risk Stratification of Gastrointestinal Stromal Tumors. \u003cem\u003eJ. Pers. Med.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e (5), 717 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, J. et al. Development of a Malignancy Potential Binary Prediction Model Based on Deep Learning for the Mitotic Count of Local Primary Gastrointestinal Stromal Tumors. \u003cem\u003eKorean J. Radiol.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e (3), 344\u0026ndash;353 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNing, Z. et al. Pattern Classification for Gastrointestinal Stromal Tumors by Integration of Radiomics and Deep Convolutional Features. \u003cem\u003eIEEE J. Biomed. Health Inf.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e (3), 1181\u0026ndash;1191 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T. et al. Radiomics nomogram for predicting the malignant potential of gastrointestinal stromal tumours preoperatively. \u003cem\u003eEur. Radiol.\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e (3), 1074\u0026ndash;1082 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, L. et al. Computed tomography-based radiomics model for discriminating the risk stratification of gastrointestinal stromal tumors. \u003cem\u003eRadiol. Med.\u003c/em\u003e \u003cb\u003e125\u003c/b\u003e (5), 465\u0026ndash;473 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoher, D., Liberati, A., Tetzlaff, J. \u0026amp; Altman, D. G. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. \u003cem\u003ePLoS Med.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e (7), e1000097 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoons, K. G. et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist.\u003cem\u003ePLoS Med\u003c/em\u003e, \u003cb\u003e11\u003c/b\u003e(10):e1001744 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWolff, R. F. et al. PROBAST: A tool to assess the risk of bias and applicability of prediction model studies. \u003cem\u003eAnn. Intern. Med.\u003c/em\u003e \u003cb\u003e170\u003c/b\u003e (1), 51\u0026ndash;58 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, L. Y. et al. Prognostic Significance of Tumor Size in 2405 Patients With Gastric Cancer: A Retrospective Cohort Study. \u003cem\u003eMed. (Baltim).\u003c/em\u003e \u003cb\u003e94\u003c/b\u003e (50), e2288 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo, P. et al. Prognostic value of tumor size in gastric cancer: an analysis of 2,379 patients. \u003cem\u003eTumour Biol.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e (2), 1027\u0026ndash;1035 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng, J. et al. Tumor size as a recommendable variable for accuracy of the prognostic prediction of gastric cancer: a retrospective analysis of 1,521 patients. \u003cem\u003eAnn. Surg. Oncol.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e (2), 565\u0026ndash;572 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSung, C. O. et al. Tumor size predicts survival in mucinous gastric carcinoma. \u003cem\u003eJ. Surg. Oncol.\u003c/em\u003e \u003cb\u003e106\u003c/b\u003e (6), 757\u0026ndash;764 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, C. Y., Shen, J. G., Shen, J. Y., Chen, W. J. \u0026amp; Wang, L. B. Ulcer size as a novel indicator marker is correlated with prognosis of ulcerative gastric cancer. \u003cem\u003eDig. Surg.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e (4), 312\u0026ndash;316 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBera, K., Braman, N., Gupta, A., Velcheti, V. \u0026amp; Madabhushi, A. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. \u003cem\u003eNat. Rev. Clin. Oncol.\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e (2), 132\u0026ndash;146 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang, B. et al. Deep learning applications in breast cancer histopathological imaging: diagnosis, treatment, and prognosis. \u003cem\u003eBreast Cancer Res.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e (1), 137 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLipkova, J. et al. Artificial intelligence for multimodal data integration in oncology. \u003cem\u003eCancer Cell.\u003c/em\u003e \u003cb\u003e40\u003c/b\u003e (10), 1095\u0026ndash;1110 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, F. A., Li, Y. \u0026amp; Zeng, T. Deep Learning of radiology-genomics integration for computational oncology: A mini review. \u003cem\u003eComput. Struct. Biotechnol. J.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, 2708\u0026ndash;2716 (2024).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Gastrointestinal stromal tumors, High malignant potential, Risk prediction model, Systematic review, Meta-analysis","lastPublishedDoi":"10.21203/rs.3.rs-5382250/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5382250/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGastrointestinal stromal tumors (GIST) present diverse biological behaviors and varying malignancy risks, necessitating early detection and effective risk stratification. Machine learning (ML) models are emerging as valuable tools for assessing the malignant potential of GIST. This systematic review and meta-analysis evaluate the efficacy of various ML models in predicting GIST malignancy. We conducted a comprehensive literature search in PubMed, EMBASE, Web of Science, and the Cochrane Library, adhering to PRISMA guidelines, up to April 22, 2024. After article selection, we extracted essential data and performed meta-analysis to aggregate the c-index, sensitivity, and specificity. The risk of bias was assessed using the PROBAST framework. Our analysis included 12 studies involving 20 ML models and 2,859 patients. Tumor size emerged as the most significant predictor. The pooled c-index was 0.89 (training) and 0.87 (test), with sensitivities of 0.85 and 0.82, and specificities of 0.89 and 0.75, respectively. Two studies had high bias risk, while ten had low bias, although overall applicability was considered low due to inadequate data sources. ML models demonstrate strong diagnostic capabilities in predicting GIST malignancy, with the Logistic Regression model performing best. Key predictive factors included tumor size, necrosis, ulceration, and shape regularity. Future models should integrate impactful disease predictors to enhance clinical utility.\u003c/p\u003e","manuscriptTitle":"Machine learning models in predicting high malignant potential in gastrointestinal stromal tumors: a systematic review and meta-analysis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-19 11:42:37","doi":"10.21203/rs.3.rs-5382250/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"47b80a2d-a3ca-48db-b4df-434a59935abe","owner":[],"postedDate":"December 19th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":41053868,"name":"Health sciences/Gastroenterology/Gastrointestinal diseases"},{"id":41053869,"name":"Biological sciences/Computational biology and bioinformatics"}],"tags":[],"updatedAt":"2025-04-07T14:45:08+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-19 11:42:37","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5382250","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5382250","identity":"rs-5382250","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}