Objective:To develop and validate machine learning-based radiomics models using preoperative CT images for individualized prediction of mitotic index (MI) in patients with gastrointestinal stromal tumors (GIST).Methods:The study was a case-control study. The data of 348 GIST patients confirmed by pathology were retrospectively collected from two independent medical centers: the Affiliated Hospital of Qingdao University (center 1) and Shandong Provincial Hospital Affiliated to Shandong First Medical University (center 2), covering the period from January 2013 to June 2018. Patients from center 1 were divided into a training cohort (176 cases) and an internal validation cohort (75 cases) at a ratio of 7∶3 using random sampling. Patients from center 2 served as an independent external validation cohort (97 cases). The primary endpoint was MI, categorized into high MI (145 cases) and low MI (203 cases) groups. Radiomic features were extracted from the portal venous phase images of preoperative contrast-enhanced CT scans. Five machine learning algorithms, including logistic regression, support vector machine, random forest, decision tree, and extreme gradient boosting (XGBoost),were employed to construct MI prediction models. The optimal model was identified using receiver operating characteristic curves. An individualized prediction model was developed by integrating the the optimal machine learning model combined with selected independent clinical factors, and the importance of features was visualized using Shapley Additive Explanation (SHAP) analysis. Patients were followed up, and Kaplan-Meier curves along with log-rank tests were used to evaluate recurrence-free survival (RFS) differences between the predicted high MI and low MI groups.Results:Among the five constructed machine learning models, the XGBoost model demonstrated the best predictive performance, with area under the curve (AUC) of 0.809 (95% CI 0.738-0.872), 0.693 (95% CI 0.571-0.809), and 0.718 (95% CI 0.605-0.822) in the training cohort, internal validation cohort, and external validation cohort, respectively. An individualized prediction model combining the XGBoost model with independent clinical factors (tumor location and tumor size) was developed. The model achieved AUC of 0.843 (95% CI 0.785-0.899), 0.791 (95% CI 0.680-0.894), and 0.777 (95% CI 0.678-0.861) in the training cohort, internal validation cohort, and external validation cohort, respectively. SHAP analysis indicated that radiomic features had the highest predictive impact. In both the training cohort and internal validation cohort, the RFS of patients predicted to be in the high MI group was lower than that of the low MI group, with statistically significant differences ( χ2=14.58, 9.52, both P<0.001). However, there was no statistically significant difference in RFS in the external validation set ( χ2=6.18, P=0.080). Conclusions:The optimal XGBoost model based on radiomic features extracted from preoperative portal venous phase CT images, when combined with clinical factors, can effectively predict the MI of GIST patients.

Objective:To develop an artificial intelligence (AI)-based automatic scoring model for magnetic resonance imaging-detected extramural vascular invasion (AI-mrEMVI) and evaluate its performance and prognostic value in patients with rectal cancer.Methods:In this multicenter retrospective cohort study, a total of 2 501 rectal cancer patients from seven centers between November 2012 and December 2020 were included and divided into completely independent training ( n=1 830) and validation ( n=671) cohorts. A nnUNet-based AI-mrEMVI scoring model was constructed. Manual mrEMVI scores assigned by two radiologists served as the reference standard for accessing the accuracy of the AI-mrEMVI scoring. Kaplan-Meier survival analysis and Cox regression were used to evaluate the prognostic stratification ability of the AI-mrEMVI scores. The concordance index (C-index) was calculated to evaluate prognostic performance. Results:In the validation cohort, the manual mrEMVI scores were 0-2 in 425 patients (63.3%), 3 in 89 (13.4%), and 4 in 157 (23.4%). The AI-mrEMVI model identified 0-2 in 375 patients (55.9%), 3 in 95 (14.2%), and 4 in 201 (30.0%), with an overall accuracy of 81.1% (544/671, 95% CI 77.9%-84.0%). The 3-year disease-free survival (DFS) rates for patients with AI-mrEMVI scores of 0-2, 3, and 4 were 85.2%, 70.0%, and 58.2%, respectively, and the 5-year overall survival (OS) rates were 87.2%, 81.6%, and 62.6%, respectively (DFS: χ2=48.74, P<0.001; OS: χ2=30.04, P<0.001). Multivariable Cox regression showed that for DFS, AI-mrEMVI scores of 3 and 4 were associated with hazard ratios ( HR) of 1.75 (95% CI 1.11-2.77, P=0.016) and 2.65 (95% CI 1.86-3.78, P<0.001), respectively. For OS, an AI-mrEMVI score of 4 was associated with an HR of 2.56 (95% CI 1.62-4.03, P<0.001). The C-index values of the AI-mrEMVI scoring model for predicting DFS and OS were 0.647 (95% CI 0.608-0.686) and 0.650 (95% CI 0.598-0.702), respectively. Conclusion:The proposed AI-mrEMVI automatic scoring model demonstrated high diagnostic accuracy and performed favorably in predicting DFS and OS prognostic risk in patients with rectal cancer.

Medical imaging large-scale models demonstrate broad application prospects in the field of tumor diagnosis and treatment.Their powerful high-dimensional feature extraction and data analysis capabilities have brought revolutionary breakthroughs to precision oncology,driving the transformation of diagnostic and therapeutic paradigms.However,current research in this field still faces numerous challenges and technical bot-tlenecks.Based on the research background of artificial intelligence(AI)large-scale models,this article sys-tematically reviews the current research status of medical imaging large-scale models from three key dimensions:the construction of large-scale medical imaging datasets,optimization of large-scale model algorithms,and com-putational resource requirements.Furthermore,it elaborates on the application scenarios of these models in pre-cision oncology and provides a forward-looking perspective on their future development.The aim is to offer prac-tical guidance for advancing precision diagnosis and treatment of tumors.

Medical imaging large-scale models demonstrate broad application prospects in the field of tumor diagnosis and treatment.Their powerful high-dimensional feature extraction and data analysis capabilities have brought revolutionary breakthroughs to precision oncology,driving the transformation of diagnostic and therapeutic paradigms.However,current research in this field still faces numerous challenges and technical bot-tlenecks.Based on the research background of artificial intelligence(AI)large-scale models,this article sys-tematically reviews the current research status of medical imaging large-scale models from three key dimensions:the construction of large-scale medical imaging datasets,optimization of large-scale model algorithms,and com-putational resource requirements.Furthermore,it elaborates on the application scenarios of these models in pre-cision oncology and provides a forward-looking perspective on their future development.The aim is to offer prac-tical guidance for advancing precision diagnosis and treatment of tumors.

Objective:To develop and validate machine learning-based radiomics models using preoperative CT images for individualized prediction of mitotic index (MI) in patients with gastrointestinal stromal tumors (GIST).Methods:The study was a case-control study. The data of 348 GIST patients confirmed by pathology were retrospectively collected from two independent medical centers: the Affiliated Hospital of Qingdao University (center 1) and Shandong Provincial Hospital Affiliated to Shandong First Medical University (center 2), covering the period from January 2013 to June 2018. Patients from center 1 were divided into a training cohort (176 cases) and an internal validation cohort (75 cases) at a ratio of 7∶3 using random sampling. Patients from center 2 served as an independent external validation cohort (97 cases). The primary endpoint was MI, categorized into high MI (145 cases) and low MI (203 cases) groups. Radiomic features were extracted from the portal venous phase images of preoperative contrast-enhanced CT scans. Five machine learning algorithms, including logistic regression, support vector machine, random forest, decision tree, and extreme gradient boosting (XGBoost),were employed to construct MI prediction models. The optimal model was identified using receiver operating characteristic curves. An individualized prediction model was developed by integrating the the optimal machine learning model combined with selected independent clinical factors, and the importance of features was visualized using Shapley Additive Explanation (SHAP) analysis. Patients were followed up, and Kaplan-Meier curves along with log-rank tests were used to evaluate recurrence-free survival (RFS) differences between the predicted high MI and low MI groups.Results:Among the five constructed machine learning models, the XGBoost model demonstrated the best predictive performance, with area under the curve (AUC) of 0.809 (95% CI 0.738-0.872), 0.693 (95% CI 0.571-0.809), and 0.718 (95% CI 0.605-0.822) in the training cohort, internal validation cohort, and external validation cohort, respectively. An individualized prediction model combining the XGBoost model with independent clinical factors (tumor location and tumor size) was developed. The model achieved AUC of 0.843 (95% CI 0.785-0.899), 0.791 (95% CI 0.680-0.894), and 0.777 (95% CI 0.678-0.861) in the training cohort, internal validation cohort, and external validation cohort, respectively. SHAP analysis indicated that radiomic features had the highest predictive impact. In both the training cohort and internal validation cohort, the RFS of patients predicted to be in the high MI group was lower than that of the low MI group, with statistically significant differences ( χ2=14.58, 9.52, both P<0.001). However, there was no statistically significant difference in RFS in the external validation set ( χ2=6.18, P=0.080). Conclusions:The optimal XGBoost model based on radiomic features extracted from preoperative portal venous phase CT images, when combined with clinical factors, can effectively predict the MI of GIST patients.

Objective:To develop an artificial intelligence (AI)-based automatic scoring model for magnetic resonance imaging-detected extramural vascular invasion (AI-mrEMVI) and evaluate its performance and prognostic value in patients with rectal cancer.Methods:In this multicenter retrospective cohort study, a total of 2 501 rectal cancer patients from seven centers between November 2012 and December 2020 were included and divided into completely independent training ( n=1 830) and validation ( n=671) cohorts. A nnUNet-based AI-mrEMVI scoring model was constructed. Manual mrEMVI scores assigned by two radiologists served as the reference standard for accessing the accuracy of the AI-mrEMVI scoring. Kaplan-Meier survival analysis and Cox regression were used to evaluate the prognostic stratification ability of the AI-mrEMVI scores. The concordance index (C-index) was calculated to evaluate prognostic performance. Results:In the validation cohort, the manual mrEMVI scores were 0-2 in 425 patients (63.3%), 3 in 89 (13.4%), and 4 in 157 (23.4%). The AI-mrEMVI model identified 0-2 in 375 patients (55.9%), 3 in 95 (14.2%), and 4 in 201 (30.0%), with an overall accuracy of 81.1% (544/671, 95% CI 77.9%-84.0%). The 3-year disease-free survival (DFS) rates for patients with AI-mrEMVI scores of 0-2, 3, and 4 were 85.2%, 70.0%, and 58.2%, respectively, and the 5-year overall survival (OS) rates were 87.2%, 81.6%, and 62.6%, respectively (DFS: χ2=48.74, P<0.001; OS: χ2=30.04, P<0.001). Multivariable Cox regression showed that for DFS, AI-mrEMVI scores of 3 and 4 were associated with hazard ratios ( HR) of 1.75 (95% CI 1.11-2.77, P=0.016) and 2.65 (95% CI 1.86-3.78, P<0.001), respectively. For OS, an AI-mrEMVI score of 4 was associated with an HR of 2.56 (95% CI 1.62-4.03, P<0.001). The C-index values of the AI-mrEMVI scoring model for predicting DFS and OS were 0.647 (95% CI 0.608-0.686) and 0.650 (95% CI 0.598-0.702), respectively. Conclusion:The proposed AI-mrEMVI automatic scoring model demonstrated high diagnostic accuracy and performed favorably in predicting DFS and OS prognostic risk in patients with rectal cancer.

Background::Artificial intelligence (AI) technology represented by deep learning has made remarkable achievements in digital pathology, enhancing the accuracy and reliability of diagnosis and prognosis evaluation. The spatial distribution of CD3 + and CD8 + T cells within the tumor microenvironment has been demonstrated to have a significant impact on the prognosis of colorectal cancer (CRC). This study aimed to investigate CD3 CT (CD3 + T cells density in the core of the tumor [CT]) prognostic ability in patients with CRC by using AI technology. Methods::The study involved the enrollment of 492 patients from two distinct medical centers, with 358 patients assigned to the training cohort and an additional 134 patients allocated to the validation cohort. To facilitate tissue segmentation and T-cells quantification in whole-slide images (WSIs), a fully automated workflow based on deep learning was devised. Upon the completion of tissue segmentation and subsequent cell segmentation, a comprehensive analysis was conducted.Results::The evaluation of various positive T cell densities revealed comparable discriminatory ability between CD3 CT and CD3-CD8 (the combination of CD3 + and CD8 + T cells density within the CT and invasive margin) in predicting mortality (C-index in training cohort: 0.65 vs. 0.64; validation cohort: 0.69 vs. 0.69). The CD3 CT was confirmed as an independent prognostic factor, with high CD3 CT density associated with increased overall survival (OS) in the training cohort (hazard ratio [HR] = 0.22, 95% confidence interval [CI]: 0.12–0.38, P <0.001) and validation cohort (HR = 0.21, 95% CI: 0.05–0.92, P = 0.037). Conclusions::We quantify the spatial distribution of CD3 + and CD8 + T cells within tissue regions in WSIs using AI technology. The CD3 CT confirmed as a stage-independent predictor for OS in CRC patients. Moreover, CD3 CT shows promise in simplifying the CD3-CD8 system and facilitating its practical application in clinical settings.

Primary liver cancer is a common malignant digestive system tumor, with hepatocellular carcinoma being the most common pathological type. Radiomics significantly boosts the efficiency of predictions by accurately capturing the intrinsically heterogeneous features of tumors that are difficult to discern with the human eye in imaging images. This article outlines the background and concepts of radiomics, introduces its latest research progress in various aspects, such as diagnosis and differential diagnosis, prediction of pathological molecular subtypes, efficacy evaluation, and survival prediction, and further discusses its limitations and prospects in HCC.

Objective:To investigate the current funding landscape of medical artificial intelligence (AI) projects in National Natural Science Foundation of China (NSFC) from 2015 to 2019.Methods:From 2015 to 2019, AI-related projects in NSFC Medical Science Department were collected. Comprehensive analysis was performed in the projects information including year, title, supporting institution, fund type, research findings, etc.Results:NSFC has funded a total of 278 projects related to artificial intelligence, with the total funding amount of 139 million yuan. The number of projects and the funding amount were increasing year by year. Among these, 90% (249/278) were general programs and young scientist funds; 53% (148/278) of the projects were regionally distributed in Beijing, Shanghai and Guangdong; 66% (184/278) of the projects were imaging-related researches; the projects mainly focused on diseases with high incidence in China, including neoplastic diseases, cardiovascular and nervous system diseases.Conclusion:The AI-related projects funded by NSFC are characterized by rapid growth in number and fund amounts, wide coverage of disciplines, and diverse types of research diseases. However, the unbalanced distribution of regions, research fields, and supporting institutions demands more attention in future.

Objective: To evaluate the effect of height (HT), total body weight (TBW), body mass index (BMI), lean body weight (LBW), body surface area (BSA) and blood volume (BV) on aortic and liver contrast enhancement during upper abdominal contrast-enhanced CT scans. Methods: One hundred and thirteen enrolled patients underwent upper abdominal multiphase contrast-enhanced CT scans. The enhancement (ΔHU) of aorta in hepatic arterial phase and liver parenchyma in portal venous phase were measured and calculated. The ΔHU values difference of aorta and liver parenchyma in subgroups between males and females, TBW<60 kg and TBW≥60 kg, BMI<25 kg/m² and BMI≥25 kg/m² were compared. To evaluate the effect of the patient′s body parameters on aortic and hepatic enhancement, we performed simple linear regression analyses between the change in CT numbers per gram of iodine (ΔHU/gI) at aorta and liver and each of the following: HT, TBW, BMI, LBW, BSA, and BV. Pearson and t test were used. Results: The mean ΔHU values of aorta and liver were significantly lower at males than that of at females (P<0.05). They were higher at TBW<60 kg patients than at TBW≥60 kg patients(P<0.05), and the mean ΔHU values of the liver at BMI<25 kg/m² patients were significantly higher than at BMI≥25 kg/m² patients(P<0.05). The proportion of the ΔHU values at liver less than 50 HU was higher at males (18.3%, 11/60) than at females (7.5%, 4/53). The most obvious negative correlation coefficients were found between the ΔHU/gI of aorta and LBW at hepatic arterial phase (r=-0.559, P<0.01), and between the ΔHU/gI of liver and BSA at portal venous phase (r=-0.680, P<0.01). Conclusion LBW or BSA could be the alternative body index to TBW for the calculation of personalized iodine dose protocol at aortic and liver enhanced CT scans.