Objective To observe the effect of self-formulated Tongluo Yizhi formula on memory deficits and endoplasmic reticulum stress (ERS) in rats with vascular dementia (VD) and its mechanism. Methods Fifty SPF-grade Wistar rats were randomly divided into sham surgery group,model group,TLYZF low-dose (12.5 mg/kg/d) group,TLYZF medium dose (25 mg/kg/d) group and TLYZF high-dose (50 mg/kg/d) group,with 10 rats in each group. Except for the Sham group,other rats were subjected to modified bilateral carotid artery ligation to construct VD models. After 4 weeks of TLYZF intervention,Morris water maze experiment was used to evaluate the learning and memory abilities of rats.Hematoxylin eosin staining was used to detect the pathological structure of the hippocampus.Nissl staining was used to detect hippocampal neuronal damage.In situ end transfer enzyme labeling staining was used to detect hippocampal neuronal apoptosis.Enzyme-linked immunosorbent assay was used to detect the content of inflammatory factors in hippocampal tissue.Protein immunoblotting was used to detect apoptosis of hippocampal tissue cells and the expression of ERS related proteins.Results Compared with the Sham group,the average escape latency of the model group was prolonged(t=14.059),the number of crossing the platform was reduced(t=8.534),the hippocampal neurons were disorderly arranged,the morphology was fuzzy,the cytoplasmic vacuolization,the nuclear pyknosis,the inflammatory cell infiltration increased,the number of Nissl bodies decreased(t=17.131),and the neuronal apoptosis rate increased(t=17.701). The contents of tumor necrosis factor-α (TNF-α),interleukin-1 β (IL-1β) and IL-6 increased(t=6.541,6.957,10.014),the expression of Bcl-2 associated X protein (Bax),cleaved caspase-3,glucose-regulated protein 78 (GRP78),phosphorylated endoplasmic reticulum kinase (p-PERK)/PERK,phosphorylated eukaryotic initiation factor 2α (p-eIF2α)/eIF2α,activating transcription factor 4 (ATF4) and CCAAT/enhancer binding protein (C/EBP) homologous protein (CHOP) increased(t=13.548～76.468),while the expression of B-cell lymphoma-2 (Bcl-2) proteinwas decreased (t=39.691),and the difference were statistically significant (all P<0.05),respectively. Compared with the model group,the average escape latency of TLYZF-treated rats was shortened(t=2.476,7.266,11.306),the number of platform crossing was increased(t=2.187,4.471,6.932),the hippocampal neurons were relatively regular,the nucleus was clear,the inflammatory cell infiltration was significantly reduced,the content of Nissl bodies was increased(t=4.359,9.477,11.449),the apoptosis rate of neurons was decreased(t=3.631,6.145,7.580),the contents of TNF-α,IL-1β and IL-6 were decreased (t=2.382～8.293),the expression of Bax(t=4.696,16.250,20.250),cleaved caspase-3(t=15.205,27.000,26.833),GRP78(t=5.918,13.139,13.741),p-PERK/PERK(t=13.416,14.230,17.889),p-eIF2α/eIF2α(t=20.152,39.346,50.750),ATF4(t=12.093,24.395,22.946)and CHOP protein((t=21.592,19.207,22.136)decreased,while Bcl-2 protein increased(t=7.474,20.761,35.350),the differences were statistically significant (all P<0.05),respectively,and the high-dose group had the best therapeutic effect.Conclusion Self-formulated Tongluo Yizhi formula could reduce the apoptosis and inflammatory injury of hippocampal neurons in VD rats,and its mechanism may be related to inhibiting ERS mediated by PERK.

Objective:To explore the effect on cardiac remodeling of combining high-intensity interval training with metformin in cases of type 1 diabetes mellitus (T1DM) and its possible mechanisms.Methods:Sixty male Sprague-Dawley rats were randomly divided into control, model, exercise, drug and combination groups with twelve per group. T1DM was induced in all except the control group using a single intraperitoneal injection of streptozotocin (60mg/kg of body weight). The control and model groups were then housed without intervention, while the drug and exercise groups underwent metformin gavage (300mg/kg/day) or 8 weeks of high-intensity interval training (3 sessions/week), respectively. The combination group receiving both. After the training, body weight, a cardiac index (heart weight/body weight), and fasting blood glucose (FBG) were measured in all of the rats. Echocardiography assessed cardiac structure and function, and myocardial tissue was collected for HE or Masson staining to measure cardiomyocyte width and collagen volume fraction respectively. Protein expression and the levels of atrial natriuretic peptide (ANP), brain natriuretic peptide (BNP), transforming growth factor-β (TGF-β), collagen (Col), peroxisome proliferator-activated receptor γ coactivator-1α (PGC-1α), phosphorylated AMP-activated protein kinase (p-AMPK), ryanodine receptor (RyR), and sarcoplasmic reticulum calcium ATPase (SERCA) were determined using western blotting.Results:Compared with the control group, the model group showed significantly increased cardiac indices, FBG levels, and left ventricular end-systolic and end-diastolic diameters. But they had significantly decreased body weight, left ventricular fractional shortening and ejection fraction. Compared with the control group, there was a significant increase in the cardiomyocyte width and collagen volume fraction, as well as in ANP, BNP, TGF-β, and Col levels in the model group, along with downregulated PGC-1α, p-AMPK, RyR, and SERCA protein expression. Compared with the model group, the exercise, drug, and combination groups exhibited varying degrees of improvement in all these indicators, with the most pronounced effects in the combination group.Conclusions:High-intensity interval training combined with metformin promotes cardiac remodeling in T1DM rats, outperforming either intervention alone. The mechanism may involve minimizing myocardial fibrosis and pathological hypertrophy, restoring mitochondrial function and intracellular calcium homeostasis.

Objective:To explore the effect on cardiac remodeling of combining high-intensity interval training with metformin in cases of type 1 diabetes mellitus (T1DM) and its possible mechanisms.Methods:Sixty male Sprague-Dawley rats were randomly divided into control, model, exercise, drug and combination groups with twelve per group. T1DM was induced in all except the control group using a single intraperitoneal injection of streptozotocin (60mg/kg of body weight). The control and model groups were then housed without intervention, while the drug and exercise groups underwent metformin gavage (300mg/kg/day) or 8 weeks of high-intensity interval training (3 sessions/week), respectively. The combination group receiving both. After the training, body weight, a cardiac index (heart weight/body weight), and fasting blood glucose (FBG) were measured in all of the rats. Echocardiography assessed cardiac structure and function, and myocardial tissue was collected for HE or Masson staining to measure cardiomyocyte width and collagen volume fraction respectively. Protein expression and the levels of atrial natriuretic peptide (ANP), brain natriuretic peptide (BNP), transforming growth factor-β (TGF-β), collagen (Col), peroxisome proliferator-activated receptor γ coactivator-1α (PGC-1α), phosphorylated AMP-activated protein kinase (p-AMPK), ryanodine receptor (RyR), and sarcoplasmic reticulum calcium ATPase (SERCA) were determined using western blotting.Results:Compared with the control group, the model group showed significantly increased cardiac indices, FBG levels, and left ventricular end-systolic and end-diastolic diameters. But they had significantly decreased body weight, left ventricular fractional shortening and ejection fraction. Compared with the control group, there was a significant increase in the cardiomyocyte width and collagen volume fraction, as well as in ANP, BNP, TGF-β, and Col levels in the model group, along with downregulated PGC-1α, p-AMPK, RyR, and SERCA protein expression. Compared with the model group, the exercise, drug, and combination groups exhibited varying degrees of improvement in all these indicators, with the most pronounced effects in the combination group.Conclusions:High-intensity interval training combined with metformin promotes cardiac remodeling in T1DM rats, outperforming either intervention alone. The mechanism may involve minimizing myocardial fibrosis and pathological hypertrophy, restoring mitochondrial function and intracellular calcium homeostasis.

Objective:To explore the effects of bone mineral density (BMD) on postoperative cage subsidence in patients undergoing anterior cervical discectomy and fusion (ACDF) in different regions of the vertebrae.Methods:The study is a retrospective case-control analysis. The clinical and imaging data of 164 cervical spondylosis patients who underwent ACDF at Department of Spine Surgery, the Third Hospital of Hebei Medical University between January 2021 and June 2024 were retrospectively reviewed. Data from 147 patients (230 intervertebral spaces) constituted the analysis set, including 80 males and 67 females, with an age of (54.5±11.2) years (range: 32 to 81 years). Patients were grouped based on postoperative cage subsidence: 74 patients were included in the subsidence group, and 73 patients were included in the non-subsidence group (grouped by patients); 99 intervertebral spaces were included in the subsidence group, and 131 intervertebral spaces were included in the non-subsidence group (grouped by intervertebral space). Cage subsidence was defined as a≥3 mm loss of intervertebral height at the operated level on lateral radiographs during follow-up. Cervical BMD was assessed using Hounsfield Units (HU) values obtained from CT images. Independent sample t-test or Chi-squared test were used to compare baseline characteristics between groups. Multivariate Logistic regression analyzed the influence of HU values at different locations on cage subsidence. Receiver operating characteristic (ROC) curve analysis calculated the area under the curve (AUC) to evaluate the predictive value of HU values for cage subsidence and determine optimal thresholds. The Delong test compared the predictive differences for subsidence among HU values from different vertebral locations at the surgical fixation levels. A validation set comprising clinical data from 17 single-level surgery patients was used to further verify the accuracy of the established thresholds for predicting cage subsidence. Results:In the analysis set, there were no statistically significant differences in age, sex, body mass index, or underlying diseases between patients in the subsidence group and the non-subsidence group (all P>0.05). The HU values of subsided intervertebral spaces were lower than those of non-subsided intervertebral spaces (upper vertebra: 360.1±86.4 vs. 301.7±93.3, t=4.899, P<0.01; lower vertebra: 328.8±83.6 vs. 282.5±88.1, t=4.062, P<0.01; endplate of the upper vertebra: 604.7±150.9 vs. 521.6±125.3, t=4.446, P<0.01; endplate of the lower vertebra: 554.4±157.9 vs. 502.8±139.0, t=2.582, P=0.010). ROC curves showed that HU values at different locations of the surgical level all had predictive value for cage subsidence (upper vertebra: AUC=0.702, P<0.01; lower vertebra: AUC=0.667, P<0.01; endplate of the upper vertebra: AUC=0.657, P<0.01; endplate of the lower vertebra:AUC=0.610, P<0.01). The optimal thresholds for predicting cage subsidence for the upper vertebral body, lower vertebral body, endplate of the upper vertebra, and endplate of the lower vertebra were 325.0, 247.1, 533.1, and 547.4, respectively. However, the differences in predictive value among HU values from different vertebral locations were not statistically significant( P>0.05). In the validation set, the sensitivity and specificity of the HU value of upper vertebra for predicting cage subsidence were 6/7 and 9/10, respectively; for the lower vertebra, they were 5/7 and 9/10; for the endplate of the upper vertebra, they were 6/7 and 8/10; for the endplate of the lower vertebra, they were 5/7 and 8/10. Conclusion:The BMD of different parts of the vertebral body has potential predictive value for cage subsidence after ACDF surgery.

Objective:To develop and validate artificial intelligence (AI) models based on deep learning for precise boundary identification of early gastric cancer (EGC) in narrow-band imaging (NBI) and near focus narrow-band imaging (NF-NBI) endoscopic images.Methods:Endoscopic submucosal dissection (ESD) images from 282 patients diagnosed as having EGC by postoperative pathology at the Department of Gastroenterology, the First Affiliated Hospital of Soochow University were retrospectively collected from February 2016 to June 2024. The images were randomly divided into the training set and the validation set at an approximate 8∶2 ratio. In the NBI modality, 980 images from 171 patients were used for training, 235 images from 61 patients were used for validation. In the NF-NBI modality, 1 273 images from 128 patients were used for training, and 373 images from 35 patients were used for validation. This study trained a total of six convolutional neural network (CNN) models: two independent CNN1 models, two independent CNN2 models, and two fused CNN3 models. Using expert-delineated EGC boundaries based on post-ESD pathological findings as the gold standard, the intersection over union (IOU) value of the CNN3 models was compared against junior (<5 years experience), mid-level (5-10 years), and senior (>10 years) endoscopists.Results:In NBI validation set, the IOU value of CNN3 model was significantly higher than that of junior (0.732 VS 0.489, Z=11.528, P<0.001) and mid-level endoscopists (0.732 VS 0.521, Z=11.184, P<0.001). However, no significant difference was observed between CNN3 model and senior endoscopists (0.732 VS 0.739, Z=0.593, P=0.554). Similarly, in NF-NBI validation set, CNN3 model outperformed junior (0.757 VS 0.537, Z=15.944, P<0.001) and mid-level endoscopists (0.757 VS 0.597, Z=9.722, P<0.001), while matching senior endoscopists (0.757 VS 0.769, Z=0.854, P=0.394). Conclusion:The fused CNN3 model achieves senior expert-level accuracy in delineating EGC boundaries in both NBI and NF-NBI images, demonstrating potential to assist less-experienced endoscopists in precise identification of EGC boundaries.

Objective To observe the effect of self-formulated Tongluo Yizhi formula on memory deficits and endoplasmic reticulum stress (ERS) in rats with vascular dementia (VD) and its mechanism. Methods Fifty SPF-grade Wistar rats were randomly divided into sham surgery group,model group,TLYZF low-dose (12.5 mg/kg/d) group,TLYZF medium dose (25 mg/kg/d) group and TLYZF high-dose (50 mg/kg/d) group,with 10 rats in each group. Except for the Sham group,other rats were subjected to modified bilateral carotid artery ligation to construct VD models. After 4 weeks of TLYZF intervention,Morris water maze experiment was used to evaluate the learning and memory abilities of rats.Hematoxylin eosin staining was used to detect the pathological structure of the hippocampus.Nissl staining was used to detect hippocampal neuronal damage.In situ end transfer enzyme labeling staining was used to detect hippocampal neuronal apoptosis.Enzyme-linked immunosorbent assay was used to detect the content of inflammatory factors in hippocampal tissue.Protein immunoblotting was used to detect apoptosis of hippocampal tissue cells and the expression of ERS related proteins.Results Compared with the Sham group,the average escape latency of the model group was prolonged(t=14.059),the number of crossing the platform was reduced(t=8.534),the hippocampal neurons were disorderly arranged,the morphology was fuzzy,the cytoplasmic vacuolization,the nuclear pyknosis,the inflammatory cell infiltration increased,the number of Nissl bodies decreased(t=17.131),and the neuronal apoptosis rate increased(t=17.701). The contents of tumor necrosis factor-α (TNF-α),interleukin-1 β (IL-1β) and IL-6 increased(t=6.541,6.957,10.014),the expression of Bcl-2 associated X protein (Bax),cleaved caspase-3,glucose-regulated protein 78 (GRP78),phosphorylated endoplasmic reticulum kinase (p-PERK)/PERK,phosphorylated eukaryotic initiation factor 2α (p-eIF2α)/eIF2α,activating transcription factor 4 (ATF4) and CCAAT/enhancer binding protein (C/EBP) homologous protein (CHOP) increased(t=13.548～76.468),while the expression of B-cell lymphoma-2 (Bcl-2) proteinwas decreased (t=39.691),and the difference were statistically significant (all P<0.05),respectively. Compared with the model group,the average escape latency of TLYZF-treated rats was shortened(t=2.476,7.266,11.306),the number of platform crossing was increased(t=2.187,4.471,6.932),the hippocampal neurons were relatively regular,the nucleus was clear,the inflammatory cell infiltration was significantly reduced,the content of Nissl bodies was increased(t=4.359,9.477,11.449),the apoptosis rate of neurons was decreased(t=3.631,6.145,7.580),the contents of TNF-α,IL-1β and IL-6 were decreased (t=2.382～8.293),the expression of Bax(t=4.696,16.250,20.250),cleaved caspase-3(t=15.205,27.000,26.833),GRP78(t=5.918,13.139,13.741),p-PERK/PERK(t=13.416,14.230,17.889),p-eIF2α/eIF2α(t=20.152,39.346,50.750),ATF4(t=12.093,24.395,22.946)and CHOP protein((t=21.592,19.207,22.136)decreased,while Bcl-2 protein increased(t=7.474,20.761,35.350),the differences were statistically significant (all P<0.05),respectively,and the high-dose group had the best therapeutic effect.Conclusion Self-formulated Tongluo Yizhi formula could reduce the apoptosis and inflammatory injury of hippocampal neurons in VD rats,and its mechanism may be related to inhibiting ERS mediated by PERK.

Objective:To study the influence of image panel of different source image distances(SID)of beam central axis of Portal Dosimetry(PD)dose verification system on the verification of volumetric modulated arc therapy(VMAT)plan,so as to provide reference for dose verification in selecting the appropriate SID position and evaluation criteria.Methods:From July 2022 to May 2024,10 patients,who received radiotherapy on tumor at different-site in Specialized Medical Center of the Chinese People's Armed Police Force,were selected.All patients underwent VMAT.Ten simulated dose verification plans were established according to the data of patients,with a total of 10 arcs.The non-crystalline silicon image panel was calibrated at 0,-30,-40,-50 and-60 cm from the isocenter on the beam central axis,and the verification plans of PD system were established.Dose verification was carried out using electronic portal imaging device(EPID)(a-Si EPID)of non-crystalline silicon image panel that was onboard by accelerator.The γ pass rates under the standards of 3%/3 mm and 3%/2 mm were compared and analyzed.Results:Under the evaluation standard of 3%/3 mm,the γ pass rates of the collected doses were(98.93±0.81)%,(94.77±9.00)%,(92.18±11.34)%,(89.72±12.54)%and(87.05±13.06)%when the SID values were respectively 100,130,140,150 and 160 cm,which were all higher than these under the evaluation standard of 3%/2 mm,and the differences were statistically significant(t=2.75,3.44,3.59,4.16,6.47,P<0.05).There were significant differences between the average γ pass rate of the collected dose under evaluation standards of 3%/3 mm and 3%/2 mm when SID was 100cm,and the 100%of γ pass rate(t=-3.21,-3.62,P<0.05),while the differences of the obtained γ pass rates among other SID positions were not statistically significant(P>0.05).Conclusion:Under the evaluation standards of 3%/3 mm and 3%/2 mm,the γ pass rate of the collected dose at SID 100 cm is the highest.The γ pass rate of the collected dose at SID 130 cm can meet the basically clinical requirements.In EPID dose verification,the collection for dose should be prioritized for selection when SID was 100 cm,and the γ pass rate of dose verification is higher,the influence of interference is less at this time,which verification results have higher credibility.

Objective:To explore the effects of bone mineral density (BMD) on postoperative cage subsidence in patients undergoing anterior cervical discectomy and fusion (ACDF) in different regions of the vertebrae.Methods:The study is a retrospective case-control analysis. The clinical and imaging data of 164 cervical spondylosis patients who underwent ACDF at Department of Spine Surgery, the Third Hospital of Hebei Medical University between January 2021 and June 2024 were retrospectively reviewed. Data from 147 patients (230 intervertebral spaces) constituted the analysis set, including 80 males and 67 females, with an age of (54.5±11.2) years (range: 32 to 81 years). Patients were grouped based on postoperative cage subsidence: 74 patients were included in the subsidence group, and 73 patients were included in the non-subsidence group (grouped by patients); 99 intervertebral spaces were included in the subsidence group, and 131 intervertebral spaces were included in the non-subsidence group (grouped by intervertebral space). Cage subsidence was defined as a≥3 mm loss of intervertebral height at the operated level on lateral radiographs during follow-up. Cervical BMD was assessed using Hounsfield Units (HU) values obtained from CT images. Independent sample t-test or Chi-squared test were used to compare baseline characteristics between groups. Multivariate Logistic regression analyzed the influence of HU values at different locations on cage subsidence. Receiver operating characteristic (ROC) curve analysis calculated the area under the curve (AUC) to evaluate the predictive value of HU values for cage subsidence and determine optimal thresholds. The Delong test compared the predictive differences for subsidence among HU values from different vertebral locations at the surgical fixation levels. A validation set comprising clinical data from 17 single-level surgery patients was used to further verify the accuracy of the established thresholds for predicting cage subsidence. Results:In the analysis set, there were no statistically significant differences in age, sex, body mass index, or underlying diseases between patients in the subsidence group and the non-subsidence group (all P>0.05). The HU values of subsided intervertebral spaces were lower than those of non-subsided intervertebral spaces (upper vertebra: 360.1±86.4 vs. 301.7±93.3, t=4.899, P<0.01; lower vertebra: 328.8±83.6 vs. 282.5±88.1, t=4.062, P<0.01; endplate of the upper vertebra: 604.7±150.9 vs. 521.6±125.3, t=4.446, P<0.01; endplate of the lower vertebra: 554.4±157.9 vs. 502.8±139.0, t=2.582, P=0.010). ROC curves showed that HU values at different locations of the surgical level all had predictive value for cage subsidence (upper vertebra: AUC=0.702, P<0.01; lower vertebra: AUC=0.667, P<0.01; endplate of the upper vertebra: AUC=0.657, P<0.01; endplate of the lower vertebra:AUC=0.610, P<0.01). The optimal thresholds for predicting cage subsidence for the upper vertebral body, lower vertebral body, endplate of the upper vertebra, and endplate of the lower vertebra were 325.0, 247.1, 533.1, and 547.4, respectively. However, the differences in predictive value among HU values from different vertebral locations were not statistically significant( P>0.05). In the validation set, the sensitivity and specificity of the HU value of upper vertebra for predicting cage subsidence were 6/7 and 9/10, respectively; for the lower vertebra, they were 5/7 and 9/10; for the endplate of the upper vertebra, they were 6/7 and 8/10; for the endplate of the lower vertebra, they were 5/7 and 8/10. Conclusion:The BMD of different parts of the vertebral body has potential predictive value for cage subsidence after ACDF surgery.

Objective:To study the influence of image panel of different source image distances(SID)of beam central axis of Portal Dosimetry(PD)dose verification system on the verification of volumetric modulated arc therapy(VMAT)plan,so as to provide reference for dose verification in selecting the appropriate SID position and evaluation criteria.Methods:From July 2022 to May 2024,10 patients,who received radiotherapy on tumor at different-site in Specialized Medical Center of the Chinese People's Armed Police Force,were selected.All patients underwent VMAT.Ten simulated dose verification plans were established according to the data of patients,with a total of 10 arcs.The non-crystalline silicon image panel was calibrated at 0,-30,-40,-50 and-60 cm from the isocenter on the beam central axis,and the verification plans of PD system were established.Dose verification was carried out using electronic portal imaging device(EPID)(a-Si EPID)of non-crystalline silicon image panel that was onboard by accelerator.The γ pass rates under the standards of 3%/3 mm and 3%/2 mm were compared and analyzed.Results:Under the evaluation standard of 3%/3 mm,the γ pass rates of the collected doses were(98.93±0.81)%,(94.77±9.00)%,(92.18±11.34)%,(89.72±12.54)%and(87.05±13.06)%when the SID values were respectively 100,130,140,150 and 160 cm,which were all higher than these under the evaluation standard of 3%/2 mm,and the differences were statistically significant(t=2.75,3.44,3.59,4.16,6.47,P<0.05).There were significant differences between the average γ pass rate of the collected dose under evaluation standards of 3%/3 mm and 3%/2 mm when SID was 100cm,and the 100%of γ pass rate(t=-3.21,-3.62,P<0.05),while the differences of the obtained γ pass rates among other SID positions were not statistically significant(P>0.05).Conclusion:Under the evaluation standards of 3%/3 mm and 3%/2 mm,the γ pass rate of the collected dose at SID 100 cm is the highest.The γ pass rate of the collected dose at SID 130 cm can meet the basically clinical requirements.In EPID dose verification,the collection for dose should be prioritized for selection when SID was 100 cm,and the γ pass rate of dose verification is higher,the influence of interference is less at this time,which verification results have higher credibility.

Objective:To develop and validate artificial intelligence (AI) models based on deep learning for precise boundary identification of early gastric cancer (EGC) in narrow-band imaging (NBI) and near focus narrow-band imaging (NF-NBI) endoscopic images.Methods:Endoscopic submucosal dissection (ESD) images from 282 patients diagnosed as having EGC by postoperative pathology at the Department of Gastroenterology, the First Affiliated Hospital of Soochow University were retrospectively collected from February 2016 to June 2024. The images were randomly divided into the training set and the validation set at an approximate 8∶2 ratio. In the NBI modality, 980 images from 171 patients were used for training, 235 images from 61 patients were used for validation. In the NF-NBI modality, 1 273 images from 128 patients were used for training, and 373 images from 35 patients were used for validation. This study trained a total of six convolutional neural network (CNN) models: two independent CNN1 models, two independent CNN2 models, and two fused CNN3 models. Using expert-delineated EGC boundaries based on post-ESD pathological findings as the gold standard, the intersection over union (IOU) value of the CNN3 models was compared against junior (<5 years experience), mid-level (5-10 years), and senior (>10 years) endoscopists.Results:In NBI validation set, the IOU value of CNN3 model was significantly higher than that of junior (0.732 VS 0.489, Z=11.528, P<0.001) and mid-level endoscopists (0.732 VS 0.521, Z=11.184, P<0.001). However, no significant difference was observed between CNN3 model and senior endoscopists (0.732 VS 0.739, Z=0.593, P=0.554). Similarly, in NF-NBI validation set, CNN3 model outperformed junior (0.757 VS 0.537, Z=15.944, P<0.001) and mid-level endoscopists (0.757 VS 0.597, Z=9.722, P<0.001), while matching senior endoscopists (0.757 VS 0.769, Z=0.854, P=0.394). Conclusion:The fused CNN3 model achieves senior expert-level accuracy in delineating EGC boundaries in both NBI and NF-NBI images, demonstrating potential to assist less-experienced endoscopists in precise identification of EGC boundaries.