Deep learning versus endoscopists for nasopharyngeal carcinoma detection: data and analysis code

Aggregate diagnostic accuracy data and reproducible R statistical code for a revised systematic review of nasopharyngeal endoscopy AI. Includes 10 primary studies, 18 extractable algorithm records within selected evaluation sets, individual-reader comparison data, 34 extracted statistical functions, reference outputs and independent-directory reproduction checks. Primary bivariate GLMM, leave-one-out analysis, reader comparisons and secondary multilevel analysis are provided. All nine numerical output files were reproduced byte-for-byte in the reference environment. Image/frame-level and patient-level observations are explicitly distinguished. Publisher PDFs, the full Shiny App, plotting code and unconfirmed quality assessments are excluded. See README for modeling assumptions, limitations and software requirements. Version 1.1.0 adds five-study overall physician comparisons, with two individual-reader and three approximate group-average comparators evaluated on the same image sets as AI. Each study and its image denominator enter once; correlations unavailable from aggregate reporting are not recovered.