ObjectiveThe essence of syndrome manifestation recognition in traditional Chinese medicine (TCM) is to infer the body’s latent pathogenesis state from clinical observational information, rather than to perform simple label matching. However, previous studies have largely modeled this task as syndrome pattern classification within a fixed label space, which does not adequately reflect the cognition process of TCM syndrome differentiation centered on pathogenesis reasoning, and is also insufficient to capture the openness, semantic variability, and cross-disease reusability of syndrome manifestation expression. This study aimed to investigate whether introducing pathogenesis reasoning chain-of-thought (PR-CoT) supervision into large language models (LLMs) could improve the quality and cognitive consistency of syndrome manifestation recognition and support cross-disease transfer. MethodsSyndrome manifestation recognition was formulated as a conditional generation task under the framework of clinical observational information (X)→pathogenesis structure (Z)→syndrome pattern output (Y), where Z serves as an explicit intermediate structural variable linking the clinical evidence and syndrome judgment. Within this framework, a PR-CoT-supervised dataset for syndrome manifestation recognition was constructed based on medical case records of spleen-stomach disorders. After preprocessing, information extraction, manual proofreading, and data cleaning, the dataset comprised 4 800 training cases, 400 development cases, and 400 test cases. Each sample was annotated with a structured PR-CoT consisting of three progressive levels: clinical information summarization, comprehensive pathogenesis analysis, and syndrome pattern output. Supervised fine-tuning was conducted on open-source LLMs, with an end-to-end model serving as the baseline. Qwen3-32B was used as the primary experimental model, and Qwen3-14B as the scale comparison model. A progressive multidimensional evaluation framework was further established, comprising a structural parsing level, a semantic similarity level, and an expert blind review level. At the structural parsing level, syndrome pattern expressions were decomposed into structural elements and evaluated using Precision, Recall, F1 score, and Jaccard similarity. At the semantic similarity level, independent LLMs scored the theoretical proximity between predicted and reference syndrome patterns. At the expert blind review level, three TCM experts independently evaluated model outputs on two dimensions: syndrome differentiation consistency and terminology standardization of syndrome patterns. In addition, zero-shot cross-disease transfer evaluation was conducted on gynecological and heart-system disorder test sets. ResultsAt the structural parsing level, PR-CoT supervision did not lead to a stable improvement in the element-wise overlap of syndrome pattern structural components. Compared with the corresponding baselines, neither Qwen3-32B nor Qwen3-14B showed consistent advantages in structural matching metrics after the introduction of PR-CoT supervision. In contrast, at the semantic similarity level, PR-CoT supervision produced stable positive gains across different model scales and evaluation systems. The average semantic score of Qwen3-32B increased from 6.425 8 in the baseline model to 6.585 0 after PR-CoT supervision, and that of Qwen3-14B increased from 5.870 0 to 5.964 2. At the expert blind review level, the overall score of Qwen3-32B (PR-CoT) was 7.026 0±0.107 7, higher than 6.416 3±0.288 9 for its baseline. In zero-shot cross-disease testing, the PR-CoT model still showed advantages in semantic evaluation and expert evaluation on both gynecological and heart-system disorder test sets, indicating a certain degree of transferability. ConclusionThe benefits of PR-CoT supervision are mainly reflected in TCM semantic consistency and clinical plausibility, rather than in improved hard matching of structural elements. These findings support understanding syndrome manifestation recognition as a process of generating and expressing latent pathogenesis structures, rather than as a classification task within a traditional fixed label space. By introducing pathogenesis reasoning as an explicit intermediate structure into the modeling process and combining it with a progressive multidimensional evaluation framework, this study provides a methodological pathway for intelligent TCM syndrome differentiation that integrates theoretical alignment, interpretability, and multi-level evaluation.

ObjectiveThe essence of syndrome manifestation recognition in traditional Chinese medicine (TCM) is to infer the body’s latent pathogenesis state from clinical observational information, rather than to perform simple label matching. However, previous studies have largely modeled this task as syndrome pattern classification within a fixed label space, which does not adequately reflect the cognition process of TCM syndrome differentiation centered on pathogenesis reasoning, and is also insufficient to capture the openness, semantic variability, and cross-disease reusability of syndrome manifestation expression. This study aimed to investigate whether introducing pathogenesis reasoning chain-of-thought (PR-CoT) supervision into large language models (LLMs) could improve the quality and cognitive consistency of syndrome manifestation recognition and support cross-disease transfer. MethodsSyndrome manifestation recognition was formulated as a conditional generation task under the framework of clinical observational information (X)→pathogenesis structure (Z)→syndrome pattern output (Y), where Z serves as an explicit intermediate structural variable linking the clinical evidence and syndrome judgment. Within this framework, a PR-CoT-supervised dataset for syndrome manifestation recognition was constructed based on medical case records of spleen-stomach disorders. After preprocessing, information extraction, manual proofreading, and data cleaning, the dataset comprised 4 800 training cases, 400 development cases, and 400 test cases. Each sample was annotated with a structured PR-CoT consisting of three progressive levels: clinical information summarization, comprehensive pathogenesis analysis, and syndrome pattern output. Supervised fine-tuning was conducted on open-source LLMs, with an end-to-end model serving as the baseline. Qwen3-32B was used as the primary experimental model, and Qwen3-14B as the scale comparison model. A progressive multidimensional evaluation framework was further established, comprising a structural parsing level, a semantic similarity level, and an expert blind review level. At the structural parsing level, syndrome pattern expressions were decomposed into structural elements and evaluated using Precision, Recall, F1 score, and Jaccard similarity. At the semantic similarity level, independent LLMs scored the theoretical proximity between predicted and reference syndrome patterns. At the expert blind review level, three TCM experts independently evaluated model outputs on two dimensions: syndrome differentiation consistency and terminology standardization of syndrome patterns. In addition, zero-shot cross-disease transfer evaluation was conducted on gynecological and heart-system disorder test sets. ResultsAt the structural parsing level, PR-CoT supervision did not lead to a stable improvement in the element-wise overlap of syndrome pattern structural components. Compared with the corresponding baselines, neither Qwen3-32B nor Qwen3-14B showed consistent advantages in structural matching metrics after the introduction of PR-CoT supervision. In contrast, at the semantic similarity level, PR-CoT supervision produced stable positive gains across different model scales and evaluation systems. The average semantic score of Qwen3-32B increased from 6.425 8 in the baseline model to 6.585 0 after PR-CoT supervision, and that of Qwen3-14B increased from 5.870 0 to 5.964 2. At the expert blind review level, the overall score of Qwen3-32B (PR-CoT) was 7.026 0±0.107 7, higher than 6.416 3±0.288 9 for its baseline. In zero-shot cross-disease testing, the PR-CoT model still showed advantages in semantic evaluation and expert evaluation on both gynecological and heart-system disorder test sets, indicating a certain degree of transferability. ConclusionThe benefits of PR-CoT supervision are mainly reflected in TCM semantic consistency and clinical plausibility, rather than in improved hard matching of structural elements. These findings support understanding syndrome manifestation recognition as a process of generating and expressing latent pathogenesis structures, rather than as a classification task within a traditional fixed label space. By introducing pathogenesis reasoning as an explicit intermediate structure into the modeling process and combining it with a progressive multidimensional evaluation framework, this study provides a methodological pathway for intelligent TCM syndrome differentiation that integrates theoretical alignment, interpretability, and multi-level evaluation.

Objective A targeted nanopeptides(VEGF/TIE-2 targeted nanopeptides,VTN)that simultaneously inhibits vascu-lar endothelial growth factor(VEGF)/tyrosine kinase with immunoglobulin-like and EGF-like domains-2(TIE-2)signaling pathways were designed and synthesized,and explore its inhibitory effect on angiogenesis in renal clear cell carcinoma(ccRCC).Methods VTN and non-self-assembling control VTN-C were prepared by solid-phase peptide synthesis technology,and the molecular structures of VTN and VTN-C were analyzed by electrospray ionization mass spectrometry(ESI-MS).The CCK-8 method was used to evaluate the effect of VTN on the cell viability of human umbilical vein endothelial cells(HUVEC).The cell scratch assay,Transwell invasion assay and angiogenesis assay were used to detect the inhibitory effects of VTN on migration,invasion and angiogenesis of HUVEC.Western blot was used to detect the effect of VTN on the phosphorylation of downstream proteins of VEGF and TIE-2 signaling pathways.A 786-O cell mouse model was established,and the effects of VTN on tumor angiogenesis and tumor progression were observed through animal ex-periments.Results ESI-MS showed that the main charge state peaks of both synthesized VTN and VVTN-C pointed to the same molec-ular weight,which was highly consistent with the corresponding theoretical molecular mass.Immunofluorescence showed that VTN co-lo-calized with VEGF and TIE-2.VTN combined with MMP-2 could significantly inhibit the activity of HUVEC(P<0.001).The cell inva-sion rate and scratch closure rate in the VTN group were reduced by(78.30±1.35)%and(37.09±3.49)%compared those in the PBS group,respectively(P<0.001).Angiogenesis experiments showed that VTN could significantly inhibit the angiogenesis of HUVEC(P<0.001).Western blot showed that VTN significantly inhibited the phosphorylation of Akt and ERK(P<0.001).The results from animal experimentsshowed that tumor volume in the VTN group was decreased by(87.16±1.30)%compared with the control group,and the CD31-positive area was reduced(P<0.001).Conclusion VTN significantly blocks ccRCC angiogenesis and delays tumor progression by inhibiting VEGF and TIE-2 signaling pathways and downregulating Akt and ERK phosphorylation.

Objective To construct a prediction model for targeted biopsy(TB)of the prostate based on multiparameter magnetic resonance imaging(mp-MRI)and prostate-specific antigen density(PSAD)to predict the outcomes TB in patients with a score of 4-5 on the Prostate Imaging Reporting and Data System(PI-RADS).Methods Clinical data of 669 patients with PI-RADS 4-5 receiving transperineal TB in our hospital during Jan.2022 and Dec.2023 were retrospectively analyzed.The data were divided into the training set and validation set with a ratio of 2∶1.Independent predictors of TB results were identified with univariate and multivariate logistic regression to construct a formula for the prediction model.A prediction model was subsequently constructed and validated using the validation set to assess its efficacy and predictive performance with the area under the receiver operating characteristic curve(AUC).The relative importance of each independent predictor in the formula was analyzed.Results Univariate and multivariate logistic regression analyses showed that age,total number of lesions,histological location,PI-RADS score and PSAD were significantly associated with the TB outcomes(P＜0.05)and could be used as independent predictors,with PI-RADS score and PSAD making the highest contribution to outcome prediction,accounting for 27.59％and 37.58％,respectively.The training set had an AUC of 0.840(95％CI:0.800-0.881),which was more predictive than other single predictors,and the high-risk group based on the optimal threshold of 0.833 increased the positive biopsy rate from 79.3％to 94.4％.The validation set had an AUC of 0.865(95％CI:0.810-0.920),and the high-risk group based on the optimal threshold of 0.594 increased the positive biopsy rate from 80.0％to 96.2％.Conclusion The prediction model has good predictive ability for lesions with PI-RADS 4-5,which can significantly improve the positive detection rate and reduce a large number of unnecessary systematic puncture.

Objective:To establish the minimum number of negative lymph nodes (nLN) required for patients undergoing gastrectomy.Methods:This was a retrospective cohort study with inclusion criteria as follows: (1) radical gastrectomy; (2) histologically confirmed adenocarcinoma; (3) complete tumor staging information; and (4) known number of lymph nodes harvested. The exclusion criteria were: (1) other concurrent malignant tumors; (2) metastatic or recurrent gastric cancer; (3) initial surgery performed at another hospital; (4) preoperative neoadjuvant therapy; (5) distant metastasis; and (6) incomplete clinical data or follow-up information. Based on the above criteria, a total of 11 167 patients with gastric adenocarcinoma who underwent radical subtotal gastrectomy (RSG) or radical total gastrectomy (RTG) in the Department of Gastric Surgery, Fudan University Shanghai Cancer Center between January 1, 2000, and December 31, 2022, were included in the study. Among them, there were 7 596 cases in the RSG group and 3 571 cases in the RTG group. Restricted cubic spline (RCS) analysis was used to determine the ideal threshold for nLN for RSG and RTG patients. Survival analysis was conducted using Kaplan-Meier (KM) curves and log-rank tests, and propensity score matching (PSM) was utilized to balance parameters between two groups. Furthermore, subgroup analysis was conducted for RSG patients based on tumor location (upper, middle and lower) to determine the minimum number of nLN in each subgroup.Results:For patients who underwent RSG, the mean number of nLN was 21.9, with a median of 21. RCS analysis showed that more than 21 nLN was associated with better survival. Moreover, both pre- and post-PSM analysis confirmed that patients with nLN ≥21 had better survival benefits compared to those with nLN <21 (overall survival [OS]: P<0.001 before PSM, P=0.013 after PSM; disease-free survival [DFS]: P<0.001 before PSM, P=0.013 after PSM). For patients who underwent RTG, the mean number of nLN was 23.5, with a median of 22. Here RCS analysis indicated that more than 22 nLN was associated with better postoperative survival in RTG patients, and both pre- and post-PSM analysis confirmed that patients with nLN ≥22 had better survival benefits compared to those with nLN<22 (OS: P<0.001 both before and after PSM; DFS: P<0.001 both before and after PSM). Subgroup analysis showed that for RSG patients with tumor located in the upper part, having ≥17 nLN (OS: both P<0.001), and for RSG patients with tumor located in the middle and lower part, having ≥22 nLN (OS: both P<0.001), were associated with better prognoses. Conclusions:For patients who receive RSG, the minimal number of nLN is ideally ≥21 (upper ≥17, middle and lower ≥22). Similarly, for patients who receive RTG, the minimum number of nLN ideally is 22.

Objective A targeted nanopeptides(VEGF/TIE-2 targeted nanopeptides,VTN)that simultaneously inhibits vascu-lar endothelial growth factor(VEGF)/tyrosine kinase with immunoglobulin-like and EGF-like domains-2(TIE-2)signaling pathways were designed and synthesized,and explore its inhibitory effect on angiogenesis in renal clear cell carcinoma(ccRCC).Methods VTN and non-self-assembling control VTN-C were prepared by solid-phase peptide synthesis technology,and the molecular structures of VTN and VTN-C were analyzed by electrospray ionization mass spectrometry(ESI-MS).The CCK-8 method was used to evaluate the effect of VTN on the cell viability of human umbilical vein endothelial cells(HUVEC).The cell scratch assay,Transwell invasion assay and angiogenesis assay were used to detect the inhibitory effects of VTN on migration,invasion and angiogenesis of HUVEC.Western blot was used to detect the effect of VTN on the phosphorylation of downstream proteins of VEGF and TIE-2 signaling pathways.A 786-O cell mouse model was established,and the effects of VTN on tumor angiogenesis and tumor progression were observed through animal ex-periments.Results ESI-MS showed that the main charge state peaks of both synthesized VTN and VVTN-C pointed to the same molec-ular weight,which was highly consistent with the corresponding theoretical molecular mass.Immunofluorescence showed that VTN co-lo-calized with VEGF and TIE-2.VTN combined with MMP-2 could significantly inhibit the activity of HUVEC(P<0.001).The cell inva-sion rate and scratch closure rate in the VTN group were reduced by(78.30±1.35)%and(37.09±3.49)%compared those in the PBS group,respectively(P<0.001).Angiogenesis experiments showed that VTN could significantly inhibit the angiogenesis of HUVEC(P<0.001).Western blot showed that VTN significantly inhibited the phosphorylation of Akt and ERK(P<0.001).The results from animal experimentsshowed that tumor volume in the VTN group was decreased by(87.16±1.30)%compared with the control group,and the CD31-positive area was reduced(P<0.001).Conclusion VTN significantly blocks ccRCC angiogenesis and delays tumor progression by inhibiting VEGF and TIE-2 signaling pathways and downregulating Akt and ERK phosphorylation.

Objective:To explore the prognosis after radical resection for stage Ⅰ gastric cancer and the application value of adjuvant chemotherapy.Methods:The retrospective cohort study was conducted. The clinicopathological data of 3 353 patients with stage Ⅰ gastric cancer who were admitted to Fudan University Shanghai Cancer Center from January 2000 to December 2022 were collected. There were 2 369 males and 984 females, aged 60(range, 21-91) years. All patients underwent radical R 0 resection. Observation indicators: (1) clinicopathological characteristics of patients; (2) influencing factors for postoperative prognosis of patients; (3) prognostic analysis of patients; (4) construction and validation of a predictive model for the efficacy of postoperative adjuvant chemotherapy. Comparison of count data between groups was conducted using the chi-square test. Univariate and multivariate analyses were performed using the Cox proportional hazards regression model. The Kaplan-Meier method was used to calculate survival rates and draw survival curves, and the Log-rank test was used for survival analysis. Based on the multivariate analysis result, a nomogram prediction model was constructed to predict survival benefit. Results:(1) Clinicopatho-logical characteristics of patients. The highly, moderately, and poorly differentiated tumors were observed in 16, 234, 396 cases of 646 patients aged <50 years and 279, 1 617, 811 cases of 2 707 pati-ents aged ≥50 years, respectively, showing a significant difference in degree of tumor differentiation between them ( P<0.05). For 297 patients in stage T1N1M0, cases aged <50 years and ≥50 years were 71 and 226, cases of males and females were 184 and 113, cases with negative and positive vascular invasion were 37 and 260, cases with negative and positive nerve invasion were 275 and 22, cases without and with postoperative adjuvant chemotherapy were 222 and 75, respectively. The above indicators for 678 patients in stage T2N0M0 105, 573, 533, 145, 517, 161, 526, 152, 563, 115, respectively. There were significant differences in the above indicators between the two groups ( P<0.05). (2) Influencing factors for postoperative prognosis of patients. Results of multivariate analysis showed that age ≥50 years, stage T2, moderately differentiated tumor, the number of lymph nodes dissected <16, positive vascular invasion, carcinoembryonic antigen (CEA) ≥5 μg/L, and CA19-9 ≥37 U/mL were independent risk factors for disease-free survival (DFS) after surgery for stage Ⅰ gastric cancer ( hazard ratio=4.600, 1.555, 1.835, 1.362, 1.451, 1.571, 2.134, 95% confidence interval as 2.806-7.541, 1.205-2.006, 1.016-3.314, 1.059-1.753, 1.057-1.993, 1.100-2.243, 1.257-3.625, P<0.05). Age ≥50 years, stage T2, the number of lymph nodes dissected <16, positive vascular invasion, CEA ≥5 μg/L, and CA19-9 ≥37 U/mL were independent risk factors for overall survival (OS) after surgery for stage Ⅰ gastric cancer ( hazard ratio=5.208, 1.597, 1.373, 1.520, 1.464, 2.356, 95% confidence interval as 3.028-8.955, 1.231-2.072, 1.060-1.777, 1.099-2.104, 1.004-2.134, 1.385-4.009, P<0.05). Postoperative adjuvant chemotherapy was an independent protective factor for both DFS and OS after surgery for stage I gastric cancer ( hazard ratio=0.361 0.297, 95% confidence interval as 0.177-0.736, 0.131-0.674, P<0.05). (3) Prognostic analysis of patients. According to the results of multi-variate analysis, among 3 353 patients, there were significant differences in 5-year DFS rate and 10-year OS rate between patients aged <50 years and ≥50 years ( P<0.05). There were significant differences in 5-year DFS rate and 10-year OS rate among patients in TNM stage ⅠA and ⅠB ( P<0.05). There were significant differences in 5-year DFS rate and 10-year OS rate among patients in stage T1N0M0, T1N1M0, T2N0M0 ( P<0.05). There were significant differences in 5-year DFS rate and 10-year OS rate among patients with the highly, moderately, and poorly differentiated tumors ( P<0.05). There were significant differences in 5-year DFS rate and 10-year OS rate among patients with the number of lymph lodes dissected <16 and ≥16 ( P<0.05). There were significant differences in 5-year DFS rate and 10-year OS rate between patients with negative and positive vascular invasion ( P<0.05). There were significant differences in 5-year DFS rate and 10-year OS rate between patients with and without postoperative adjuvant chemotherapy ( P<0.05), among patients in stage T1N0M0, T1N1M0, T2N0M0 who received no postoperative adjuvant chemotherapy ( P<0.05). For patients in stage T1N1M0, there was no significant difference in 5-year DFS rate and 10-year OS rate between patients with and without postoperative adjuvant chemotherapy ( P>0.05).Results of stratified analysis showed that for patients aged ≥ 50 years, there were significant differences in 5-year DFS rate and 10-year OS rate between patients with and without postoperative adjuvant chemotherapy ( P<0.05). For patients in stage T2N0M0, there were significant differences in 5-year DFS rate and 10-year OS rate between patients with and without postoperative adjuvant chemotherapy ( P<0.05). For patients with positive vascular invasion, there were significant differences in 5-year DFS rate and 10-year OS rate between patients with and without postoperative adjuvant chemotherapy ( P<0.05). (4) Construction and validation of a predictive model for the efficacy of adjuvant chemotherapy. A nomogram predictive model was constructed based on the multivariate analysis results of OS and used for calculating net benefits and distribution. Among the 3 096 patients without postoperative adjuvant chemotherapy, 1 009 cases had a predicted net benefit of >5%-10%, and 250 patients had a predicted net benefit >10%. The predicted survival analysis further verified that the predicted benefit of adjuvant chemotherapy was consistent with the prognosis of patients. Conclusions:Patients with age ≥50 years, stage T2 tumors, moderately differentiated tumor, the number of lymph nodes dissected <16, positive vascular invasion have worse survival prognosis postoperative. Postoperative adjuvant chemotherapy provides better prognosis in high-risk patients. Patients in stage T1N1M0 have lower recurrence and survival risks, of whom with 1 metastatic lymph node is more suitable for follow-up rather than postoperative adjuvant chemotherapy.

Objective To explore the effectiveness of an integrated laparoscopic simulation training course (best essential surgical technique training, BEST) in enhancing laparoscopic peritoneal suturing techniques in surgical residents.Methods As an integrated two-stage program, the BEST course applied basic laparoscopic training system with simple molds in phase Ⅰ training, and then adopted advanced laparoscopic training system, 3D Laparoscope and ex-vivo animal models in phase Ⅱ training. The laparoscopic suturing techniques were practiced in phase Ⅱ training. From August 2021 to July 2024, surgical residents in the second year of the national standardized training program were divided into pilot and control groups based on whether they had undergone the BEST course. Two cases of laparoscopic peritoneal suture were performed by the surgical residents under supervision in the department of gastrointestinal surgery. The operative time, quality of suture, and independent completion rate were compared between the two groups.Results A total of 33 surgical residents (19 in pilot group and 14 in control group) were included in this study, and a total of 66 cases of laparoscopic peritoneal suture were performed (38 in pilot group and 28 in control group). The operative time was significantly shorter in pilot group than that in control group (15.7 min vs. 17.5 min, P=0.025). The quality of suture was significantly better in pilot group compared to control group (P=0.023). In pilot group, all peritoneal sutures were performed by residents independently, whereas in control group, 3 cases (10.7%) were assisted by the supervisor, and the independent completion rate was different significantly (P=0.039).Conclusions The BEST course can help improve surgical residents′ laparoscopic peritoneal suturing techniques and could be promoted in the national standardized training program for surgical residents.

Cancer stem cells (CSCs) are widely acknowledged as primary mediators to the initiation and progression of tumors. The association between microbial infection and cancer stemness has garnered considerable scholarly interest in recent years. Porphyromonas gingivalis (P. gingivalis) is increasingly considered to be closely related to the development of oral squamous cell carcinoma (OSCC). Nevertheless, the role of P. gingivalis in the stemness of OSCC cells remains uncertain. Herein, we showed that P. gingivalis was positively correlated with CSC markers expression in human OSCC specimens, promoted the stemness and tumorigenicity of OSCC cells, and enhanced tumor formation in nude mice. Mechanistically, P. gingivalis increased lipid synthesis in OSCC cells by upregulating the expression of stearoyl-CoA desaturase 1 (SCD1) expression, a key enzyme involved in lipid metabolism, which ultimately resulted in enhanced acquisition of stemness. Moreover, SCD1 suppression attenuated P. gingivalis-induced stemness of OSCC cells, including CSCs markers expression, sphere formation ability, chemoresistance, and tumor growth, in OSCC cells both in vitro and in vivo. Additionally, upregulation of SCD1 in P. gingivalis-infected OSCC cells was associated with the expression of KLF5, and that was modulated by P. gingivalis-activated NOD1 signaling. Taken together, these findings highlight the importance of SCD1-dependent lipid synthesis in P. gingivalis-induced stemness acquisition in OSCC cells, suggest that the NOD1/KLF5 axis may play a key role in regulating SCD1 expression and provide a molecular basis for targeting SCD1 as a new option for attenuating OSCC cells stemness.
Porphyromonas gingivalis/pathogenicity* ; Stearoyl-CoA Desaturase/metabolism* ; Humans ; Carcinoma, Squamous Cell/pathology* ; Mouth Neoplasms/metabolism* ; Animals ; Neoplastic Stem Cells/microbiology* ; Mice, Nude ; Mice ; Nod1 Signaling Adaptor Protein/metabolism* ; Kruppel-Like Transcription Factors/metabolism* ; Cell Line, Tumor

Objective:To apply artificial intelligence（AI）in classifying ovarian tumors on ultrasound images，and compare the diagnostic results of several sonographers with varying seniority levels.Methods:A total of 645 patients diagnosed with adnexal masses via gynecological ultrasound examination at Qilu Hospital of Shandong University from January 2021 to December 2024 were enrolled. Three deep learning architectures，i.e.，Alexnet，Densenet121，and Resnet50 were developed and used to internally test the classification effectiveness of ovarian tumors，while the optimal model was selected for external testing. Two junior sonographers and two senior sonographers were recruited to independently diagnose ovarian tumors in the external test dataset. Subsequently，the benign and malignant results of the model's predictions were disclosed to each sonographer，and their revised diagnoses on the same external test data in combination with the best AI model were recorded.Results:The optimal model achieved an accuracy of 0.941，sensitivity of 0.936，and specificity of 0.944 on the internal test dataset，and maintained robust performance on the external test dataset with accuracy of 0.891，sensitivity of 0.880，and specificity of 0.907. Compared to junior sonographers，the optimal model demonstrated significantly higher sensitivity in discriminating benign from malignant ovarian tumors（0.880 vs. 0.723，0.602；all P<0.05）. No statistically significant difference was observed in diagnostic accuracy between the optimal model and senior sonographer 1（ P=0.05）. With assistance from the optimal model，junior sonographers achieved significant improvements in both sensitivity and specificity（sensitivity：0.723 vs. 0.843，0.602 vs. 0.819；specificity：0.778 vs. 0.833，0.685 vs. 0.741；all P<0.05）. Conclusions:The optimal model achieves comparable performance to that of senior sonographers in ovarian tumor classification. With model assistance，the diagnostic performance of junior sonographers is significantly improved.