Development And Validation Of A Predictive Model For The Diagnosis Of Bladder Tumors Using Narrow Band Imaging

Background:White light cystoscope(WLC)is recommended in guidelines for screening and post-treatment recurrence monitoring of bladder tumours as a common screening tool for bladder tumours.Narrow band imaging(NBI),an emerging high-resolution optical imaging technique,has been shown to offer greater advantages over WLC in the diagnosis of bladder tumours,particularly in the detection of flattened and smaller papillary lesions that are easily missed with WLC.We found that despite the current widespread use of NBI technology in bladder tumour screening,the specificity of NBI cystoscopy currently needs to be further improved due to the lack of uniform diagnostic criteria,and in the study we focused on this issue and attempted to provide a reference protocol for this.Methods:All patients included in the analysis underwent plain white light imaging(WLI)combined with NBI cystoscopy,during which we recorded bladder tumour morphology and tumour mucosal vascular characteristics in both imaging modalities.Pathological findings were obtained in all patients.Bootstrap resampling was used for internal validation of the model and data from 71 patients from four other centres were used for external validation of the model.Results:Information on 231 patients requiring cystoscopy was collected prospectively from November 2021 to November 2022,and after pre-analysis data collation,12 patients containing missing indicators were excluded.Six predictors were finally included in the model through univariate and multifactorial logistic regression analysis,which were:tumour morphology(cauliflower-like,alginate-like),bladder mucosal vascular morphology in NBI mode(punctate vessels,circumferential vessels),tumour boundary in NBI mode(clear or blurred);and patient’s haematuria symptoms.The internal validation showed that the area under curve(AUC)of the model was 0.94(95%CI:0.92-0.97),with a goodness of fit test p=0.97.The external validation showed that the AUC of the model was 0.89(95%CI:0.82-0.97),with a goodness of fit test p=0.24.Both internal and external validation showed excellent predictive performance of the model.Conclusion:In this study,an NBI-based model for predicting the nature of bladder tumours was developed,which was able to predict the risk of malignancy in the NBI mode during cystoscopy.The model was internally validated and externally validated to show excellent predictive performance and to improve the specificity of NBI microscopy compared to previous studies,with the aim of providing reference evidence for clinicians’ decisions and thus helping to promote accurate screening for bladder tumours.