Purpose To explore the value of contrast-enhanced CT habitat imaging(HI)in preoperative non-invasive visualization for predicting pathological grading of clear cell renal carcinoma(ccRCC).Materials and Methods A retrospective analysis was conducted on enhanced CT images and clinical data from 240 patients with pathologically confirmed ccRCC at Xijing Hospital,the Fourth Military Medical University from January 2020 to December 2023.All patients were randomly divided into training and test sets at a 7:3 ratio and classified into low-grade group(International Society of Urological Pathology Ⅰ-Ⅱ)and high-grade group(International Society of Urological Pathology Ⅲ-Ⅳ)based on postoperative pathology.Using wash-in and wash-out parametric maps,the tumors were segmented into three perfusion-based habitat subregions(low,medium and high)via K-means clustering,and the volume fraction of each subregion was calculated.Predictive factors were selected from habitat features and clinical variables(including sex,age,tumor size,etc.)using Logistic regression.Three models were constructed:a clinical model,a habitat imaging model and a combined clinical-habitat model.Model performance was evaluated using receiver operating characteristic curve,calibration curve and decision curve analysis.Results Habitat 3 exhibited higher wash-in and wash-out gradients compared to Habitats 1 and 2,indicating hyper perfusion.Its proportion was significantly higher in the low-grade group than in the high-grade group(Z=-7.71,-5.11,both P<0.01).Multivariate Logistic regression identified hypertension,maximum tumor diameter and platelet-to-lymphocyte ratio as independent risk factors for high-grade ccRCC,while the proportion of Habitat 3 was a protective factor(OR=0.297,95％CI 0.184-0.479).The combined clinical-habitat model demonstrated the highest predictive performance[area under the curve(AUC)=0.938],significantly outperforming the clinical model(AUC=0.801,Z=-3.832,P<0.01)and the habitat imaging model(AUC=0.895,Z=-2.157,P=0.031).Conclusion The clinical-habitat imaging model achieves the highest predictive performance for ccRCC pathological grading.Contrast-enhanced CT habitat imaging provides significant incremental value in predicting ccRCC pathological grading,showing potential to guide precision medicine in clinical practice.

Objective To construct a hypertension structured database based on Yi-9B large language model by aiming at the large amount of unstructured data generated in the process of hypertension diagnosis and treatment in order to elevate the efficiency of data management and provide the support for clinical deci-sion-making.Methods The key clinical informations of 114 369 patients with hypertension visiting in the Sec-ond Affiliated Hospital of Army Medical University during 2014-2023 were extracted.The Yi-9B large lan-guage model was used for conducting the entity identification and data structuring,and the database architec-ture was designed for statistical analysis and clinical application.Results After the database structuring process,the mean values of systolic and diastolic blood pressure were(149.98±20.55)mmHg and(86.90±13.75)mmHg,respectively.According to the classification of blood pressure level,the proportions of the nor-mal high value for high risk,very high risk of hypertension grade 1,and very high risk of hypertension grade 2 were the highest,which accounted for 20.73％,27.80％and 19.59％respectively.52.64％of the patients were complicated with heart disease,10.18％with complicating diabetes and 12.71％with complicating hy-perlipidemia.Logistic regression analysis showed that＞50-60 and＞60-70 years old was the high incidence age segment,moreover the systolic blood pressure showed an increasing trend with the age increase,reflecting the universality of hypertension in aging.This database significantly improved the efficiency of diagnosis and treatment in clinical application and realized the efficient analysis and management of data.Conclusion The hyper-tension structured database based on Yi-9B large language model effectively processes the unstructured data,significantly improves the efficiency of data extraction and management,helps to optimize the diagnosis and treatment decision-making,improves the management efficiency and provides the support for intelligent man-agement and personalized diagnosis and treatment.

Objective To construct the predictive model of pressure injury(PI)in the patients with chronic pain based on machine learning,and to analyze its accuracy and rationality,so as to provide an evidence for the predictive evaluation of clinical PI.Methods The clinical medical records data of 396 patients with chronic pain and high risk Braden scores hospitalized in a class 3A hospital of Chongqing City from March 2023 to June 2024 were retrospectively analyzed.Based on the Python3.10 programming language,the decision tree model,random forest model,linear regression model,naive Bayes model and K-Means model were con-structed,and the model performances were compared by accuracy,sensitivity,precision,F1 score and area un-der the receiver operating characteristic(ROC)curve(AUC).Results PI occurred in 35 cases with an inci-dence rate of 8.84%.Age,NRS score,pain site and pain affected sleep were the independent influencing fac-tors for the PI occurrence in the patients with chronic pain.Among 5 kinds of PI risk predictive model,the ac-curacy(0.873),sensitivity(0.874),precision(0.848),F1 score(0.844)and ROC AUC(0.81)of the ran-dom forest model were all higher than those of other models.Conclusion The random forest model has a high predictive performance for PI in the patients with chronic pain,and could be used for the screening and man-agement of high risk groups of PI in the patients with chronic pain.

This article explores the construction and practice of the"Party Building+Health Science Popularization"model,using the"Yixian Health Science Popularization Guangdong Tour"campaign conducted by Sun Yat-sen Memorial Hospital as a case study.The initiative has achieved remarkable results.Additionally,it summarizes innovative measures,as well as uni-versal and exemplary experiences,providing new insights and pathway recommendations for public hospitals to develop the"Party Building+Health Science Popularization"model.

Objective:To explore the value of non-invasive habitat imaging (HI) multi-parametric MRI (mpMRI) in predicting the risk of prostate cancer (PCa).Methods:In this cross-sectional study, 220 patients with PCa confirmed by radical prostatectomy (RP) who underwent multi-parametric MRI (mpMRI) scanning at Xijing Hospital, Air Force Military Medical University from January 2018 to May 2024 were retrospectively collected. Patients were divided into a training set (154 cases) and a test set (66 cases) by simple random sampling in a 7∶3 ratio. Based on mpMRI imaging, the apparent diffusion coefficient (ADC), perfusion fraction (f), and mean kurtosis (MK) of each voxel were integrated. The K-means clustering algorithm was used to divide the PCa target lesions into habitat subregions, generate habitat maps, and calculate the proportion of each habitat subregion in the entire lesion. According to the 2019 International Society of Urological Pathology (ISUP) guidelines, patients were categorized into a low-risk group (ISUP≤2, 65 cases) and a high-risk group (ISUP≥3, 155 cases). The RP specimens were matched with the habitat map to identify corresponding habitat subregions, and the ISUP grade of each subregion was individually evaluated to calculate the detection rate of high-risk PCa patients. The logistic regression analysis was applied to identify the independent risk factors associated with PCa risk, and the HI-clinical imaging model and clinical imaging model were constructed. The efficacy of the models was assessed using receiver operating characteristic curve.Results:Based on the optimal cluster number, the habitat was divided into three subregions. Habitat 1 had lower ADC and f values and higher MK values, while habitat 2 had the opposite characteristics, and habitat 3 was intermediate. The proportion of habitat 1 in the high-risk group was 28.8%, in the low-risk group was 8.9%. In the training set, the comparison of habitat subregions with pathological results showed that the detection rate of high-risk lesions was 66.9% (103/154) in habitat 1, 25.3% (39/154) in habitat 2, and 47.4% (73/154) in habitat 3. The logistic regression analysis indicated that the proportion of habitat 1 ( OR=3.03, 95% CI 1.77-5.18, P<0.001), prostate-specific antigen ( OR=1.66, 95% CI 1.04-2.66, P=0.034), and the prostate imaging reporting and data system score ( OR=1.65, 95% CI 1.00-2.70, P=0.048) as independent risk factors for high-risk PCa. In the training set, the area under the curve (AUC) for predicting PCa risk was 0.854 (95% CI 0.789-0.920) for the HI-clinical imaging model and 0.779 (95% CI 0.701-0.856) for the clinical imaging model. In the test set, the AUC values were 0.809 (95% CI 0.693-0.895) and 0.738 (95% CI 0.619-0.856), respectively. Conclusion:HI based on mpMRI can effectively predict the risk of PCa.

This article explores the construction and practice of the"Party Building+Health Science Popularization"model,using the"Yixian Health Science Popularization Guangdong Tour"campaign conducted by Sun Yat-sen Memorial Hospital as a case study.The initiative has achieved remarkable results.Additionally,it summarizes innovative measures,as well as uni-versal and exemplary experiences,providing new insights and pathway recommendations for public hospitals to develop the"Party Building+Health Science Popularization"model.

Objective:To explore the value of non-invasive habitat imaging (HI) multi-parametric MRI (mpMRI) in predicting the risk of prostate cancer (PCa).Methods:In this cross-sectional study, 220 patients with PCa confirmed by radical prostatectomy (RP) who underwent multi-parametric MRI (mpMRI) scanning at Xijing Hospital, Air Force Military Medical University from January 2018 to May 2024 were retrospectively collected. Patients were divided into a training set (154 cases) and a test set (66 cases) by simple random sampling in a 7∶3 ratio. Based on mpMRI imaging, the apparent diffusion coefficient (ADC), perfusion fraction (f), and mean kurtosis (MK) of each voxel were integrated. The K-means clustering algorithm was used to divide the PCa target lesions into habitat subregions, generate habitat maps, and calculate the proportion of each habitat subregion in the entire lesion. According to the 2019 International Society of Urological Pathology (ISUP) guidelines, patients were categorized into a low-risk group (ISUP≤2, 65 cases) and a high-risk group (ISUP≥3, 155 cases). The RP specimens were matched with the habitat map to identify corresponding habitat subregions, and the ISUP grade of each subregion was individually evaluated to calculate the detection rate of high-risk PCa patients. The logistic regression analysis was applied to identify the independent risk factors associated with PCa risk, and the HI-clinical imaging model and clinical imaging model were constructed. The efficacy of the models was assessed using receiver operating characteristic curve.Results:Based on the optimal cluster number, the habitat was divided into three subregions. Habitat 1 had lower ADC and f values and higher MK values, while habitat 2 had the opposite characteristics, and habitat 3 was intermediate. The proportion of habitat 1 in the high-risk group was 28.8%, in the low-risk group was 8.9%. In the training set, the comparison of habitat subregions with pathological results showed that the detection rate of high-risk lesions was 66.9% (103/154) in habitat 1, 25.3% (39/154) in habitat 2, and 47.4% (73/154) in habitat 3. The logistic regression analysis indicated that the proportion of habitat 1 ( OR=3.03, 95% CI 1.77-5.18, P<0.001), prostate-specific antigen ( OR=1.66, 95% CI 1.04-2.66, P=0.034), and the prostate imaging reporting and data system score ( OR=1.65, 95% CI 1.00-2.70, P=0.048) as independent risk factors for high-risk PCa. In the training set, the area under the curve (AUC) for predicting PCa risk was 0.854 (95% CI 0.789-0.920) for the HI-clinical imaging model and 0.779 (95% CI 0.701-0.856) for the clinical imaging model. In the test set, the AUC values were 0.809 (95% CI 0.693-0.895) and 0.738 (95% CI 0.619-0.856), respectively. Conclusion:HI based on mpMRI can effectively predict the risk of PCa.

Purpose To explore the value of contrast-enhanced CT habitat imaging(HI)in preoperative non-invasive visualization for predicting pathological grading of clear cell renal carcinoma(ccRCC).Materials and Methods A retrospective analysis was conducted on enhanced CT images and clinical data from 240 patients with pathologically confirmed ccRCC at Xijing Hospital,the Fourth Military Medical University from January 2020 to December 2023.All patients were randomly divided into training and test sets at a 7:3 ratio and classified into low-grade group(International Society of Urological Pathology Ⅰ-Ⅱ)and high-grade group(International Society of Urological Pathology Ⅲ-Ⅳ)based on postoperative pathology.Using wash-in and wash-out parametric maps,the tumors were segmented into three perfusion-based habitat subregions(low,medium and high)via K-means clustering,and the volume fraction of each subregion was calculated.Predictive factors were selected from habitat features and clinical variables(including sex,age,tumor size,etc.)using Logistic regression.Three models were constructed:a clinical model,a habitat imaging model and a combined clinical-habitat model.Model performance was evaluated using receiver operating characteristic curve,calibration curve and decision curve analysis.Results Habitat 3 exhibited higher wash-in and wash-out gradients compared to Habitats 1 and 2,indicating hyper perfusion.Its proportion was significantly higher in the low-grade group than in the high-grade group(Z=-7.71,-5.11,both P<0.01).Multivariate Logistic regression identified hypertension,maximum tumor diameter and platelet-to-lymphocyte ratio as independent risk factors for high-grade ccRCC,while the proportion of Habitat 3 was a protective factor(OR=0.297,95％CI 0.184-0.479).The combined clinical-habitat model demonstrated the highest predictive performance[area under the curve(AUC)=0.938],significantly outperforming the clinical model(AUC=0.801,Z=-3.832,P<0.01)and the habitat imaging model(AUC=0.895,Z=-2.157,P=0.031).Conclusion The clinical-habitat imaging model achieves the highest predictive performance for ccRCC pathological grading.Contrast-enhanced CT habitat imaging provides significant incremental value in predicting ccRCC pathological grading,showing potential to guide precision medicine in clinical practice.

Objective:To design and evaluate the application value of intracavitary-interstitial brachytherapy (IC-ISBT) applicator template for locally advanced cervical cancer.Methods:MRI data of 100 patients with ⅡB-ⅣA stage cervical cancer (International Federation of Gynecology and Obstetrics 2018 staging system) before and after external beam radiation therapy (EBRT) admitted to Sun Yat-sen University Cancer Center from March 2019 to September 2020 were collected. The range of primary cervical lesions was retrospectively analyzed and compared. Based on the residual mass of patients, the corresponding high-risk clinical target volume (HR-CTV) was delineated, and the IC-ISBT applicator template was designed and initially applied to cervical cancer patients. Dosimetry analysis and efficacy evaluation were compared between the applicator template-guided ( n=37) and free-hand implantation groups ( n=63). Chi-square test or Fisher exact test was performed for categorical variables, and t-test or U-test for continuous variables. Results:The median distance between the residual tumor margin (clockwise 3, 6, 9, 12 o'clock) and the center of 100 patients with ⅡB-ⅣA stage cervical cancer after EBRT was 16.5, 14.0, 17.0 and 13.0 mm, respectively. The corresponding HR-CTV was superimposed to reconstruct the three-dimensional diagram, and the cylindrical IC-ISBT applicator template with mushroom-like head was designed and manufactured: the longest and shortest diameter of the head was 35 and 20 mm, respectively; the central channel was adapted to the uterine tube, the C1-C12 channels was arranged in inner circle, and the peripheral B1-B5 and A1-A4 pin channels were expanded bilaterally. In terms of dose coverage, there was no significant difference between the HR-CTV D 90% [(635.12±22.65) vs. (635.80±25.84) cGy], bladder D 2 cm3 [(473.79±44.78) vs. (463.55±66.43) cGy)], rectum D 2 cm3 [(396.99±73.54) vs. (408.00±73.94) cGy] and sigmoid colon D 2 cm3 [(293.07±152.72) vs. (311.31±135.77) cGy] between the template-guided and free-hand implantation groups (all P>0.05), but the HR-CTV D 98% was significantly higher [(544.78±32.07) vs. (536.78±32.04) cGy, P=0.007] and the rectum D 1 cm3 and D 0.1 cm3 were significantly lower [(438.62±69.65) vs. (453.97±67.89) cGy, P=0.016; (519.46±70.67) vs. (543.82±81.24) cGy, P=0.001] in the template-guided implantation group. In addition, there was no significant difference in the complete response rate between two groups (86% vs. 83%, P>0.05). Conclusions:This IC-ISBT applicator template is reasonably designed, and the therapeutic efficacy of the template-guided implantation is equivalent to that of free-hand implantation. The dose coverage of the target area meets the clinical demand with a better protection of the organs at risk. The applicator template has the potential to be widely used as a conventional template in clinical practice as the applicator-guided implantation is convenient to operate and repeat.

Objective To investigate the dosimetric difference between manual and inverse optimization in 3-dimensional (3D) brachytherapy for gynecologic tumors. Methods This retrospective study was conducted among a total of 110 patients with gynecologic tumors undergoing intracavitary combined with interstitial brachytherapy or interstitial brachytherapy. Based on the original images, the brachytherapy plans were optimized for each patient using Gro, IPSA1, IPSA2 (with increased volumetric dose limits on the basis of IPSA1) and HIPO algorithms. The dose-volume histogram (DVH) parameters of the clinical target volume (CTV) including V200, V150, V100, D90, D98 and CI, and the dosimetric parameters D2cc, D1cc, and D0.1cc for the bladder, rectum, and sigmoid colon were compared among the 4 plans. Results Among the 4 plans, Gro optimization took the longest time, followed by HIPO, IPSA2 and IPSA1 optimization. The mean D90, D98, and V100 of HIPO plans were significantly higher than those of Gro and IPSA plans, and D90 and V100 of IPSA1, IPSA2 and HIPO plans were higher than those of Gro plans (P<0.05), but the CI of the 4 plans were similar (P>0.05). For the organs at risk (OARs), the HIPO plan had the lowest D2cc of the bladder and rectum;the bladder absorbed dose of Gro plans were significantly greater than those of IPSA1 and HIPO (P<0.05). The D2cc and D1cc of the rectum in IPSA1, IPSA2 and HIPO plans were better than Gro (P<0.05). The D2cc and D1cc of the sigmoid colon did not differ significantly among the 4 plans. Conclusion Among the 4 algorithms, the HIPO algorithm can better improve dose coverage of the target and lower the radiation dose of the OARs, and is thus recommended for the initial plan optimization. Clinically, the combination of manual optimization can achieve more individualized dose distribution of the plan.