{"id":"W4284884719","doi":"10.1002/jum.16052","title":"Human‐to‐<scp>AI</scp> Interrater Agreement for Lung Ultrasound Scoring in <scp>COVID</scp>‐19 Patients","year":2022,"lang":"en","type":"article","venue":"Journal of Ultrasound in Medicine","topic":"Ultrasound in Clinical Applications","field":"Medicine","cited_by":37,"is_retracted":false,"has_abstract":true,"ca_institutions":"University Hospital Foundation","funders":"","keywords":"Medicine; Coronavirus disease 2019 (COVID-19); Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Inter-rater reliability; 2019-20 coronavirus outbreak; Lung ultrasound; Betacoronavirus; Ultrasound; Radiology; Internal medicine; Virology; Disease","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.004502901,0.0005113701,0.001438589,0.001286499,0.0003576835,0.00005583488,0.0008776031,0.0001703451,0.000522408],"category_scores_gemma":[0.02860041,0.0004337574,0.0003373724,0.001460636,0.0002863378,0.0003090376,0.0001669791,0.002140565,0.00001797227],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001726153,"about_ca_system_score_gemma":0.0003766195,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001265478,"about_ca_topic_score_gemma":0.0001000257,"domain_scores_codex":[0.9932294,0.0003123436,0.003104496,0.0007124288,0.001691968,0.0009493298],"domain_scores_gemma":[0.9833655,0.01311941,0.001224957,0.0007266293,0.0006453845,0.0009181303],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001211195,0.002061918,0.7034176,0.0002832556,0.0002246315,0.0000593905,0.009164468,0.0008412236,0.03536728,0.0002396139,0.2479686,0.0002509196],"study_design_scores_gemma":[0.02819424,0.009642166,0.624408,0.001849129,0.0007491535,0.0009730613,0.02428576,0.0001147029,0.001146608,0.007446232,0.3008682,0.0003227534],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9894477,0.0006677341,0.002203718,0.002026353,0.001442448,0.002085914,0.00006936052,0.00003550918,0.002021337],"genre_scores_gemma":[0.9823003,0.0002169086,0.002508614,0.01041069,0.001257517,0.0003863684,0.0001763041,0.0000976937,0.002645543],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07900959,"threshold_uncertainty_score":0.9998114,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0424251902796635,"score_gpt":0.3726194353020092,"score_spread":0.3301942450223457,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}