Objective To evaluate the performance of lightweight Chinese large language models (LLMs) in answering specialized lung cancer questions, and to explore the impact of retrieval-augmented generation (RAG) on model performance. Methods Eleven lightweight Chinese LLMs with parameter sizes ranging from 7B to 32B were included. A lung cancer-specific evaluation dataset consisting of 200 questions [100 A1-type (basic knowledge) and 100 A2-type (clinical case) questions], constructed based on clinical guidelines and thoracic surgery textbooks, was used for assessment. Model performance was evaluated under two conditions (with and without RAG). Accuracy was used to assess model performance, and response latency was recorded to reflect inference efficiency. An accuracy–latency scatter plot was constructed for descriptive analysis of overall model performance. Results All models successfully completed the evaluation. With the introduction of RAG, the overall average accuracy improved from 61.68% to 76.36%. Smaller models demonstrated the most significant improvement (e.g., the accuracy of DeepSeek-7B increased from 32.50% to 60.00%, P<0.001). The average response latency increased from 12.58 s to 13.80 s. The Qwen3 series showed the best overall performance, and Qwen3-32B achieved the highest accuracy under both conditions (76.50% and 84.00%, respectively). After RAG integration, performance differences among model families were markedly reduced. Based on the accuracy-latency trade-off, Qwen3-32B achieved the best balance between accuracy and response latency under the baseline condition, whereas Qwen3-14B demonstrated superior overall performance in terms of accuracy, latency, and computational cost after RAG integration. Conclusion The integration of RAG technology improves the ability of lightweight Chinese LLMs to answer specialized lung cancer questions. Under the dual practical constraints of limited computational resources and medical data security requirements, the "lightweight model+RAG" technical framework may represent a promising deployment solution.

This study aims to investigate the anti-tumor effect and mechanism of Guiqi Yiyuan Ointment combined with cisplatin on Lewis lung carcinoma-bearing mice via the protein kinase RNA-like endoplasmic reticulum kinase(PERK)/eukaryotic translation initiation factor 2α(eIF2α)/activated transcription factor 4(ATF4)/C/EBP homologous protein(CHOP) signaling pathway. Sixty SPF-grade male C57BL/6 mice were selected and assigned into a blank group and a modeling group by the random number table method. After modeling of the Lewis lung carcinoma, the mice in the modeling group were randomized into model, cisplatin(5 mg·kg~(-1), once a week), and low-, medium-, and high-dose(1.7, 3.5, and 7.05 g·kg~(-1), respectively, once a day) Guiqi Yiyuan Ointment+cisplatin(5 mg·kg~(-1)) groups(n=10). After 14 days of continuous intervention, the spleen, thymus, and tumor samples of the mice were collected, weighed, and recorded, and the spleen index, thymus index, and tumor suppression rate were calculated. Hematoxylin-eosin(HE) staining was employed to observe the pathological changes in the tumor tissue. The morphological changes of the endoplasmic reticulum of tumor cells were observed by transmission electron microscopy. The positive expression of phosphorylated eIF2α(p-eIF2α) and ATF4 in the tumor tissue was detected by immunofluorescence. Western blot was employed to determine the protein levels of phosphorylated PERK(p-PERK), p-eIF2α, ATF4, CHOP, B-cell lymphoma-2(Bcl-2), Bcl-2-associated X protein(Bax), cyclin-dependent kinase inhibitor 1A(p21), and cyclinD1 in the tumor tissue. Real-time fluorescent quantitative PCR was employed to determine the mRNA levels of PERK, eIF2α, ATF4, CHOP, Bax, Bcl-2, p21, and cyclinD1 in the tumor tissue. Compared with the blank group, the model group showed decreases in spleen index and thymus index(P<0.05). Compared with the model group, the cisplatin group showed decreases in spleen index and thymus index(P<0.05), and the medium-and high-dose Guiqi Yiyuan Ointment+cisplatin groups presented increases in spleen index and thymus index(P<0.05). In addition, the treatment groups all showed decreased tumor mass(P<0.05), increased tumor cell lysis and nuclear rupture, widened gap between rough endoplasmic reticulum, enhanced average fluorescence intensity of p-eIF2α and ATF4(P<0.05), up-regulated protein levels of p-PERK/PERK, p-eIF2α/eIF2α, ATF4, CHOP, Bax, and p21(P<0.05), down-regulated protein and mRNA levels of Bcl-2 and cyclinD1(P<0.05), and up-regulated mRNA levels of PERK, eIF2α, ATF4, CHOP, Bax, and p21(P<0.05). Compared with the cisplatin group, the combination groups showed increases in spleen index and thymus index(P<0.05) as well as mean optical density(P<0.05), and the high-dose Guiqi Yiyuan Ointment+cisplatin group showed decreased tumor mass(P<0.05). In addition, the medium-and high-dose Guiqi Yiyuan Ointment+cisplatin groups showcased enhanced average fluorescence intensity of p-eIF2α and ATF4(P<0.05), up-regulated protein levels of p-PERK/PERK, p-eIF2α/eIF2α, ATF4, CHOP, Bax, and p21(P<0.05), down-regulated protein and mRNA levels of Bcl-2 and cyclinD1(P<0.05), and up-regulated mRNA levels of PERK, eIF2α, ATF4, CHOP, Bax, and p21(P<0.05). In conclusion, Guiqi Yiyuan Ointment combined with cisplatin can effectively inhibit the growth of Lewis lung carcinoma in mice by regulating the expression of proteins related to the PERK/eIF2α/ATF4/CHOP signaling pathway and promoting cell cycle arrest and apoptosis.
Animals ; Cisplatin/administration & dosage* ; Activating Transcription Factor 4/genetics* ; Eukaryotic Initiation Factor-2/genetics* ; eIF-2 Kinase/genetics* ; Carcinoma, Lewis Lung/pathology* ; Drugs, Chinese Herbal/administration & dosage* ; Male ; Mice ; Signal Transduction/drug effects* ; Mice, Inbred C57BL ; Transcription Factor CHOP/genetics* ; Ointments/administration & dosage* ; Humans ; Cell Proliferation/drug effects* ; Antineoplastic Agents/administration & dosage*

Objective·To develop a multi-algorithm collaborative computational biology strategy for constructing a predictive model of the peripheral morphine tolerance network and for screening high-confidence candidate targets.Methods·A murine model of morphine tolerance was established across multiple treatment time points.Bulk RNA sequencing was performed on harvested dorsal root ganglion(DRG)tissues.Using the expression matrix as a basis,a weighted gene co-expression network was constructed to identify co-expressed gene modules.Candidate genes were subsequently screened through the integration of differentially expressed genes(DEGs)with key weighted gene co-expression network modules.These candidates underwent functional annotation via Gene Ontology(GO)and Kyoto Encyclopedia of Genes and Genomes(KEGG)enrichment analyses.A protein-protein interaction(PPI)network was established,and hub genes were systematically identified using the cytoHubba algorithm.Three distinct machine learning approaches,least absolute shrinkage and selection operator(LASSO)regression,support vector machine recursive feature elimination(SVM-RFE)model,and random forest(RF)model,were strategically integrated to screen characteristic signature genes.Finally,gene set enrichment analysis(GSEA)was implemented to functionally validate both the hub and signature genes.Results·Weighted gene co-expression network analysis(WGCNA)identified 8 297 key module genes,of which 177 candidate genes overlapped with DEGs.These genes were significantly enriched in biological processes including ion channel regulation and vascular smooth muscle contraction.A combination of PPI network analysis and machine learning revealed four signature genes[actin γ2,smooth muscle(Actg2),centriolar coiled-coil protein 110(Ccp110),neural cell adhesion molecule 2(Ncam2),and selenium binding protein 1(Selenbp1)]and six hub genes[actin α2,smooth muscle(Acta2),von Willebrand factor(Vwf),cellular communication network factor 2(Ccn2),integrin β4(Itgb4),integrin α11(Itga11),and TEK receptor tyrosine kinase(Tek)]closely associated with morphine tolerance.Conclusion·In this study,we successfully constructed a multi-algorithm collaborative peripheral nerve regulation network prediction model for morphine tolerance,and screened out 10 core genes with high confidence.

Objective·To develop a multi-algorithm collaborative computational biology strategy for constructing a predictive model of the peripheral morphine tolerance network and for screening high-confidence candidate targets.Methods·A murine model of morphine tolerance was established across multiple treatment time points.Bulk RNA sequencing was performed on harvested dorsal root ganglion(DRG)tissues.Using the expression matrix as a basis,a weighted gene co-expression network was constructed to identify co-expressed gene modules.Candidate genes were subsequently screened through the integration of differentially expressed genes(DEGs)with key weighted gene co-expression network modules.These candidates underwent functional annotation via Gene Ontology(GO)and Kyoto Encyclopedia of Genes and Genomes(KEGG)enrichment analyses.A protein-protein interaction(PPI)network was established,and hub genes were systematically identified using the cytoHubba algorithm.Three distinct machine learning approaches,least absolute shrinkage and selection operator(LASSO)regression,support vector machine recursive feature elimination(SVM-RFE)model,and random forest(RF)model,were strategically integrated to screen characteristic signature genes.Finally,gene set enrichment analysis(GSEA)was implemented to functionally validate both the hub and signature genes.Results·Weighted gene co-expression network analysis(WGCNA)identified 8 297 key module genes,of which 177 candidate genes overlapped with DEGs.These genes were significantly enriched in biological processes including ion channel regulation and vascular smooth muscle contraction.A combination of PPI network analysis and machine learning revealed four signature genes[actin γ2,smooth muscle(Actg2),centriolar coiled-coil protein 110(Ccp110),neural cell adhesion molecule 2(Ncam2),and selenium binding protein 1(Selenbp1)]and six hub genes[actin α2,smooth muscle(Acta2),von Willebrand factor(Vwf),cellular communication network factor 2(Ccn2),integrin β4(Itgb4),integrin α11(Itga11),and TEK receptor tyrosine kinase(Tek)]closely associated with morphine tolerance.Conclusion·In this study,we successfully constructed a multi-algorithm collaborative peripheral nerve regulation network prediction model for morphine tolerance,and screened out 10 core genes with high confidence.

Objective:Analysis of surgical strategies for atrial functional mitral regurgitation with atrial fibrillation.Methods:Retrospective analysis of 112 patients with mitral regurgitation and atrial fibrillation between June 2017 and January 2023. Among them, 56 cases were severe atrial functional mitral regurgitation with atrial fibrillation, and the other 56 cases were degenerative mitral regurgitation with atrial fibrillation. All patients underwent maze Ⅳ procedure and mitral valve surgery. Follow up will be conducted through outpatient follow-up and telephone calls. The condition of postoperative mitral valve is obtained through echo. The postoperative cardiac rhythm is based on the patient's conscious symptoms, electrocardiogram, 24 hour dynamic electrocardiogram.Results:The comparison of preoperative basic data shows that the age, duration of atrial fibrillation, and comorbidity of patients with atrial functional mitral regurgitation are significantly higher than those in the degenerative mitral regurgitation group. All patients successfully completed the surgery. Postoperative death occurred in 2 cases in the atrial mitral regurgitation group. The causes of death were ARDS and pulmonary infection, respectively. The main postoperative complications include bleeding, low cardiac output, pulmonary infection, and acute kidney injury. During follow-up, 43 patients (79.6%) in the atrial mitral regurgitation group maintained sinus rhythm, while 49 patients (87.5%) in the degenerative group. However, there was no statistically significant difference in the Kaplan- Meier curves. In the atrial mitral regurgitation group, there were 47 cases with no mitral regurgitation, 4 cases with mild regurgitation, and 1 case with moderate regurgitation. In the degenerative group, there were 42 cases with no mitral regurgitation, 6 cases with mild regurgitation, 1 case with moderate regurgitation, and 1 case with severe regurgitation. The risk for atrial fibrillation recurrence in the atrial mitral regurgitation is related to postoperative left atrial diameter greater than 50 mm, while in the degenerative group, atrial fibrillation recurrence is related to postoperative left atrial diameter greater than 50 mm and residual mitral regurgitation. Conclusion:Mitral valve repair combined with maze Ⅳ procedure is an effective treatment for patients with severe atrial functional mitral regurgitation and atrial fibrillation. Further improving the success rate of atrial fibrillation and reducing surgical trauma will benefit patients in the future.

Objective To explore the changes of the thickness of myometrium(MT),apparent diffusion coefficient value of myometrium(ADCM),thickness of junctional zone(JZT)and apparent diffusion coefficient value of junctional zone(ADCJz)in patients with endometrial fibrosis.Methods A total of 59 patients with endometrial fibrosis and 34 healthy women(volunteer)of childbearing age were prospectively included.The JZT,ADCJz,MT and ADCM were measured.Independent samples t-test was used to compare the differences in JZT,ADCJZ,MT,and ADCM between the two groups.A combined prediction model was established using binary logistic regression analysis(combining mean JZT,mean ADCJZ,and mean MT).The efficiency of each parameter's mean value and the combined prediction model in diagnosing endometrial fibrosis was evaluated using the receiver operating characteristic(ROC)curve.Results JZT(anterior wall,posterior wall,fundus and mean;P=0.007,0.035,0.001 and＜0.001,respectively),ADCJZ(anterior wall,posterior wall,fundus and mean;all P＜0.001)and MT(anterior wall,posterior wall and mean;P=0.003,＜0.001 and 0.003,respectively)were significantly larger in patients with endometrial fibrosis compared to volunteer.Mean ADCJZ[area under the curve(AUC)=0.872]and the combined prediction model(AUC=0.953)had high value for diagnosing endometrial fibrosis.Conclusion MRI can be used for noninvasively assessing the changes of myometrium and JZ in patients with endometrial fibrosis.

AIM To study the secondary metabolites from endophytic fungi of Taxus wallichiana var.chinensis (Pilger) Florin and their biological activities.METHODS Endophytes were isolated and purified by tissue block and scribing method,the obtained strains were identified by molecular method.Macroporous resin column and silica gel column were used for separation and purification,then the structures of obtained compounds were identified by physicochemical properties and spectral data.The inhibition of phosphatidylinositol 3-kinase,phosphodiesterase and fusarium oxysporum activities were determined by chemiluminescence kinase assay kit method,resazurin cell viability assay and enzyme linked immunosorbent assay.RESULTS Eight endophytic fungi were isolated and identified as Trichoderma,Aspergillus,Gongronella,Phoma,Penicillium.Strain ZGG-5 had strong PI3K inhibitory activity with IC50 value of (0.17±0.05)μg/mL,and was identified as Gongronella butleri.Thirteen compounds were isolated from ZGG-5,and compound 13 had strong activities of inhibiting PI3K at concentration of 1 mg/mL,and the inhibitory rate was (97.8±0.51)％.CONCLUSION T.wallichiana var.chinensis contains endophytic fungi with strong activities of inhibiting PI3K.Compounds 1-12 are first isolated from G.butleri.

AIM To study the secondary metabolites from endophytic fungi of Taxus wallichiana var.chinensis (Pilger) Florin and their biological activities.METHODS Endophytes were isolated and purified by tissue block and scribing method,the obtained strains were identified by molecular method.Macroporous resin column and silica gel column were used for separation and purification,then the structures of obtained compounds were identified by physicochemical properties and spectral data.The inhibition of phosphatidylinositol 3-kinase,phosphodiesterase and fusarium oxysporum activities were determined by chemiluminescence kinase assay kit method,resazurin cell viability assay and enzyme linked immunosorbent assay.RESULTS Eight endophytic fungi were isolated and identified as Trichoderma,Aspergillus,Gongronella,Phoma,Penicillium.Strain ZGG-5 had strong PI3K inhibitory activity with IC50 value of (0.17±0.05)μg/mL,and was identified as Gongronella butleri.Thirteen compounds were isolated from ZGG-5,and compound 13 had strong activities of inhibiting PI3K at concentration of 1 mg/mL,and the inhibitory rate was (97.8±0.51)％.CONCLUSION T.wallichiana var.chinensis contains endophytic fungi with strong activities of inhibiting PI3K.Compounds 1-12 are first isolated from G.butleri.

Objective:To study the influence of Mei mini maze procedure for atrial functional mitral regurgitation.Methods:The data of 33 patients with atrial fibrillation and atrial functional mitral regurgitation from January 2017 to June 2020 were retrospectively analyzed. All patients received Mei mini maze procedure for atrial fibrillation. The procedure is carried out thoracoscopically through the left thoracic approach. The ablation of atrial fibrillation includes bilateral circumferential pulmonary vein ablation, isolation of the left atrium posterior wall, left atrial appendage resection, ablation of Marshall's ligament and autonomic ganglion, etc. Follow-up was conducted by outpatient follow-up and telephone. Postoperative heart rhythm was recorded by the patient's symptoms, electrocardiogram, 24 h holter and other examinations. Postoperative mitral valve lesions were obtained by echocardiography.Results:33 patients successfully completed the operation. There was no conversion to thoracotomy and no perioperative death. Thirty patients(90.9%) maintained sinus rhythm at discharge. Before discharge, 16 patients had no mitral regurgitation in echocardiography, 8 patients had mild mitral regurgitation, and 9 patients had moderate mitral regurgitation. Follow-up was 1-4 years after discharge, with a mean of(2.6±1.1) years. Sinus rhythm was maintained in 23 patients(69.7%). 17 patients had no mitral regurgitation, 9 had mild mitral regurgitation, 6 had moderate, and 1 had severe mitral regurgitation. The degree of regurgitation in 25 patients was reduced compared with pre-operation, 5 patients remained unchanged, and 3 patients mitral regurgitation aggravated. Unreduced atrial functional mitral regurgitation was associated with recurrence of atrial fibrillation by Cox multivariate analysis.Conclusion:This study found a close relationship between atrial fibrillation rhythm and atrial functional mitral regurgitation. Most moderate atrial functional mitral regurgitation can be alleviated by effective treatment for atrial fibrillation. It is not recommended that patients with severe atrial functional mitral regurgitation only receive treatment for atrial fibrillation.

Objective:To evaluate the value of ultrasonographic features and classification defined by the Chinese-Thyroid Imaging Reporting and Data System (C-TIRADS) in differentiating follicular thyroid carcinoma (FTC) from follicular thyroid adenoma (FTA) .Methods:Data of 87 patients who underwent thyroid nodule-related surgery at Tianjin Medical University General Hospital from Jan. 2019 to Feb. 2023 and whose postoperative pathology confirmed FTC or FTA was divided into FTC group (26 cases) and FTA group (61 cases) were retrospectively analyzed. The differences of general data, pathological diagnosis, surgical methods, ultrasonographic features of thyroid nodules and C-TIRADS classification between the two groups were compared. Multivariate Logistic regression analysis was used to screen the independent predictors related to the occurrence of FTC, a regression model was established based on the independent predictors and a nomogram was drawn. Bootstrap method was used to verify the nomogram internally, and ROC curve and calibration curve were drawn to evaluate the diagnostic efficiency and predictive ability of the nomogram.Results:There were statistically significant differences between orientation, echo, composition, calcification, margin, halo and C-TIRADS classification between groups (all P<0.05) ; maximum diameter, echotexture, blood flow, general data, pathological diagnosis and surgical methods had no statistically significant differences between groups (all P>0.05) .Single-factor logistic regression analysis showed taller-than-wide, peripheral calcification, blurred/irregular edges or external invasion, C-TIRADS 4B/4C class were candidate variables for predicting the occurrence of FTC ( OR=10.909, 19.059, 19.259, 49.333, all P<0.05). Multivariate logistic regression analysis showed that peripheral calcification, blurry/irregular edges or extra-thyroid invasion, C-TIRADS 4B/4C class were independent predictors for the occurrence of FTC ( OR=33.731, 11.620, 22.794, all P<0.05) .The nomogram of FTC occurrence was constructed based on the above three independent predictors, and the AUC was 0.920 (95% CI=0.854-0.986), which showed that the nomogram had high diagnostic efficacy, and the optimal diagnostic threshold determined by the Youden index was 0.23 (corresponding to the total score of the nomogram is about 52 points), the sensitivity was 88.5%, and the specificity was 82.0%. The calibration curve drawn by internal verification was close to the reference line, suggesting that the nomogram had a high predictive ability. Conclusions:The nomogram prediction model for FTC occurrence based on the ultrasonographic features and classification defined by C-TIRADS has high diagnostic efficiency and predictive ability, and can help in the preoperative identification of FTC and FTA.