Chinese journal of otorhinolaryngology head and neck surgery最新文献

Objective: To develop a nasopharyngeal carcinoma (NPC) diagnostic model based on foundation model transfer learning, aiming to address the limited generalization and diagnostic performance of existing models under small-sample conditions. Methods: A retrospective study was conducted using 27 362 nasopharyngeal endoscopic images from eight regional NPC centers. The images were classified into three groups: NPC, benign hyperplasia (BH), and normal nasopharynx (NOR). The data were randomly split into a training/validation set (85%) and a hold-out test set (15%). To evaluate generalization under small-sample conditions, models were trained on both the full dataset (100%) and a small subset (1%), then tested on the same test set. The model was based on BiomedCLIP, pre-trained on large medical image-text datasets and fine-tuned for classification. The performance of our fine-tuned BiomedCLIP model was systematically compared against several benchmark models, including ResNet50, ViT-Base, and the original CLIP. Performance was assessed using accuracy, the area under the receiver operating characteristic curve (AUC), specificity, and sensitivity, with attention maps used to visualize how the model made its decisions. Results: With the full training data, the BiomedCLIP model demonstrated robust performance. It achieved 95.46% (95%CI: 94.87%-96.08%) accuracy and an AUC of 0.98 (95%CI: 0.98-0.99) for distinguishing normal vs abnormal cases (NAN), and 89.92% (95%CI: 89.04%-90.78%) accuracy and an AUC of 0.90 (95%CI: 0.89-0.90) for distinguishing malignant vs non-malignant cases (MNM), significantly outperforming all comparator models. Even when trained with only 1% of the data, BiomedCLIP still maintained strong performance, with AUCs of 0.89 (95%CI: 0.88-0.90) for NAN and 0.81 (95%CI: 0.79-0.82) for MNM, demonstrating effective generalization in data-scarce scenarios. Conclusions: The endoscopic image-based auxiliary diagnostic model presented in this study accurately differentiates NPC, BH, and NOR under small-sample conditions. The model exhibits high diagnostic accuracy and robust generalization despite limited training data, highlighting its promise for clinical deployment as a screening and decision-support tool.

Objective: To evaluate HE2Signature for predicting inflammatory gene expression and postoperative outcomes in chronic rhinosinusitis with nasal polyps (CRSwNP) directly from whole slide images (WSIs). Methods: In an independent external cohort of 178 CRSwNP patients, HE2Signature was employed to analyze WSIs to predict expression of 33 inflammatory marker genes. Post-operative control was assessed (EPOS 2020). Predicted gene expression was correlated with clinical indicators and compared across control groups. LASSO regression was used to build a prediction model for post-operative outcome, compared with ControlNet. Results: The predicted expression levels of 18 of the 33 marker genes showed significant correlations with clinical indicators (P<0.05). The predicted expression of type 2 genes (e.g., POSTN, FCER2, IL-13) were significantly and positively correlated with eosinophil-related metrics and disease burden (r=0.160-0.244). Conversely, predicted expression of non-type 2 genes (e.g., CSF3, SAA1) was positively associated with tissue neutrophil counts. The uncontrolled disease group was characterized by a significant upregulation of predicted Type 2 inflammatory genes. A model based on the predicted gene signature achieved an area under the receiver operating characteristic curve (AUC) of 0.776 for discriminating uncontrolled status and 0.800 for discriminating control status. This performance was not statistically different from that of the ControlNet model (P>0.05). Conclusion: The HE2Signature model effectively predicts marker gene expression and postoperative outcomes in CRSwNP from routine histology, offering a scalable and intelligent pathway for precision medicine without requiring molecular assays.

Objectives: To investigate the pathological inflammatory features based on artificial intelligence for whole slide image (AI-WSI), and to evaluate its consistency and clinical relevance with the conventional mean of ten random high-power fields (10-HPF). Ultimately, a WSI-based pathological endotype classification for chronic rhinosinusitis with nasal polyps (CRSwNP) was established. Methods: A total of 407 CRSwNP patients admitted to the Department of Otorhinolaryngology-Head and Neck Surgery, the Third Affiliated Hospital of Sun Yat-sen University from January 2020 to December 2023 were retrospectively enrolled. The cohort included 288 males and 119 females, aged from 18 to 84 years. Quantitative analysis of inflammatory cells in the pathological sections of these patients was performed using the traditional 10 HPF method and the AI-WSI method, respectively. Subsequently, a typing system was established based on AI-WSI results using unsupervised clustering, discriminant analysis, and classification tree modeling, and the tissue components and clinical characteristics of each subtype were compared. Data analysis was conducted using SPSS 26.0 and R 4.4.2 software. Results: Significant differences were observed in the proportion of inflammatory cells between the AI-WSI and 10 HPF methods, with a Cohen's Kappa coefficient of 0.48. The existing 10 HPF-based typing criterion was not suitable for the inflammatory feature results of AI-WSI. Cluster analysis of AI-WSI data identified four distinct subtypes: eosinophil (Eos)-predominant, plasma cell (Pla)-predominant, lymphocyte (Lym)-predominant, and neutrophil (Neu)-predominant. The Eos-predominant subtype accounted for 28.26%, characterized by the highest recurrence rate (39.13%) and olfactory dysfunction. The Lym-predominant and Pla-predominant subtypes presented milder symptoms, with recurrence rates of 13.63% and 16.13%, respectively. Although the Neu-predominant subtype was associated with significant head and facial pain, it had a lower recurrence rate (11.11%). Conclusions: There are differences in the pathological inflammatory features between the traditional 10 HPF method and the AI-WSI method, and the features derived from AI-WSI are currently difficult to directly apply to existing typing criterion. This study successfully establishes a four-subtype classification system for CRSwNP based on AI-WSI, which demonstrates good stability and discriminative ability.

Objective: To investigate the response to induction chemotherapy (IC) as prognostic indicators for radiotherapy response and survival outcomes in patients with loco-regionally advanced tonsil squamous cell carcinoma (LATSCC). Methods: A retrospective analysis was conducted on 84 LATSCC patients who were admitted to the Eye & ENT Hospital of Fudan University. There were 67 males and 17 females, aged from 42 to 76 years. Among them, 55 patients (65.5%) were p16 positive and 29 patients (34.5%) were p16 negative. All patients received initial IC with TPF chemotherapy regimen (docetaxel+cisplatin+capecitabine) followed by either radiotherapy alone or concurrent chemoradiotherapy. The differences in categorical variables were compared using Chi-square test and Fisher exact test. Logistic regression was employed to identify independent factors associated with radiotherapy sensitivity. Kaplan-Meier method was utilized to calculate cumulative survival, while Log-rank test and Cox risk model were performed for univariate and multivariate analyses of overall survival (OS), locoregional relapse-free survival (LRFS), and distant metastasis free survival (DMFS). Results: After IC, the objective response rates (ORR) in p16 positive group and p16 negative group were 87.3% and 89.7%, respectively (χ²=0.103, P=0.749). Following radiotherapy, the complete response rates in p16 positive group and p16 negative group were 87.3% and 72.4%, respectively (χ²=5.058, P=0.115). Sensitivity to IC (OR=10.883, 95%CI: 2.555-45.930, P=0.001) was independently associated with complete response to radiotherapy and in the p16 positive group, the 3-year OS, LRFS, and DMFS were 90.0%, 81.5%, and 96.4%, respectively. In the p16 negative group, they were 82.1%, 89.1%, and 92.0%, with no statistically significant difference in the rates between two groups(P>0.05). Compared to IC-resistant patients, IC-sensitive patients showed significant improvements in 3-year LRFS (90.2% vs. 40.0%, χ²=19.750, P<0.001). IC resistance was identified as an independent risk factor for LRFS (HR=2.180, 95%CI=1.235-3.849, P=0.007). Conclusions: The response to IC is a significant prognostic factor for tonsil squamous cell carcinoma treated with definitive radiotherapy. Poor response to IC is associated with unsatisfactory outcomes either in p16-positive or p16-negative cancer.

Objective: To analyze the current status of diagnosis, treatment and prognostic factors of poorly differentiated thyroid carcinoma (PDTC) and anaplastic thyroid carcinoma (ATC) using a multicenter cohort database. Methods: A retrospective analysis was performed on clinical data of 882 patients with PDTC/ATC who received treatment from 14 centers between January 2016 and June 2025. There were 346 cases of PDTC and 536 cases of ATC, including 424 cases of conventional ATC (con-ATC) and 112 cases of coexisting papillary and anaplastic thyroid carcinoma (co-PTC/ATC). Demographic, clinicopathologic, and treatment and follow-up data were collected. Kaplan-Meier method was used to plot survival curves, and Cox proportional hazards regression model was applied to evaluate the factors influencing overall survival (OS). Results: This cohort included 514 females and 368 males. 74.72%(659/882) of the patients were over 55 years. At the time of diagnosis, the tumors frequently invaded vital adjacent structures, with 58.28% (514/882) of patients presenting with stage T3b-T4b disease. Distant metastases were observed in 27.66% (244/882) of patients, most commonly involving the lungs (78.28%, 191/244). The median OS of con-ATC patients was 8.0 months (95%CI: 7.0-10.0). The median OS of co-PTC/ATC patients was 36.0 months (95%CI: 18.0-65.0), that of PDTC patients was 104.0 months (95%CI: 72.0-NA). Patients who underwent surgery as the initial treatment appeared to have prolonged survival compared with those receiving systemic therapy (surgery 36.0 months (95%CI: 28.0-57.0) vs systemic therapy 12.0 months (95%CI: 9.0-17.0): HR (95%CI)=1.342 (1.100-1.637), P=0.004). Among patients who underwent R2 resection followed by adjuvant radiotherapy, OS did not differ significantly from that of patients who achieved R0 or R1 resection (R0/R1 resection 44.0 months (95%CI: 32.0-65.0) vs R2 resection followed by adjuvant radiotherapy 25.0 months (95%CI: 18.0-46.0): HR (95%CI)=0.860 (0.654-1.131), P=0.279). For patients with unresectable PDTC/ATC, combined targeted therapy and immunotherapy were associated with superior overall survival compared with other systemic treatment regimens (combined targeted therapy and immunotherapy 13.0 months (95%CI: 9.0-27.0) vs other systemic treatment regimens 9.0 months (95%CI: 6.0-16.0): HR (95%CI)=0.681 (0.495-0.936), P=0.018). Conclusions: Long-term survival is achievable in selected patients with PDTC/ATC through active multimodal treatment. When formulating treatment strategies, radical surgery should be considered as the key to cure.

Objective: To establish a predictive system for aerosol drug deposition by integrating computational fluid dynamics (CFD) and artificial intelligence (AI) modeling, and to propose optimized strategies for intranasal drug delivery in patients with chronic rhinosinusitis with nasal polyps (CRSwNP) after various Draf procedures. Methods: Post-operative 3D nasal airway models of Draf Ⅱa, Ⅱb, and Ⅲ were reconstructed based on CT data from a CRSwNP patient treated at the Second Affiliated Hospital of Xi'an Jiaotong University. CFD simulations were employed to assess the impact of inspiratory flow rates (15, 30, and 45 L/min), nebulizer oxygen flow rates (6 and 8 L/min), and particle sizes (5-50 μm) on drug deposition efficiency (DE) across three target areas (TAs): the frontal sinus (TA1), the olfactory region and ethmoid sinus (TA2), and the respiratory zone (TA3). Based on 540 sets of CFD data, XGBoost machine learning models were developed to predict regional DE and to interpret variable importance. Statistical analysis was performed using SPSS 21.0. Results: Regarding particle size, optimal DE was observed with 25 μm drug particles for TA1 (DE_max=30.19%), 15-25 μm for TA2 (27.55%), and 20-35 μm for TA3 (25.77%). Nowadays, clinically available nebulizers producing small particles (1-5 μm) generated extremely low DE in TA1 (<1%) and TA2 (<2%). Among the surgical variants, Draf Ⅲ provided the largest frontal sinus ostium with the highest airflow velocity (1.53 m/s), resulting in the highest DE of TA1 (τ=0.75, P<0.001). However, Draf IIb achieved the highest cumulative DE across all three TAs (τ=0.40, P=0.001). The XGBoost models exhibited excellent predictive performance (R²: 0.81-0.98). Feature importance analysis and SHapley Additive exPlanations (SHAP) values revealed that drug particle size was the primary determinant of cumulative DE across all three TAs (accounting for >40% of importance), while surgical procedure had a dominant influence on the DE of TA1 (accounting for >40% of importance), with the Draf Ⅲ surgery significantly promoting frontal sinus drug deposition. Conclusion: By integrating CFD and AI techniques, this study demonstrates that the Draf III procedure significantly improves frontal sinus ventilation and inhances drug deposition. Medium-sized particles (25 μm) combined with low-to-moderate inspiratory flow rates (15-30 L/min) are optimal for nasal-targeted aerosol therapy.