{"id":"W4384561684","doi":"10.1038/s41591-023-02437-x","title":"Enhancing the reliability and accuracy of AI-enabled diagnosis via complementarity-driven deferral to clinicians","year":2023,"lang":"en","type":"article","venue":"Nature Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":110,"is_retracted":false,"has_abstract":false,"ca_institutions":"Google (Canada)","funders":"","keywords":"Deferral; Complementarity (molecular biology); Reliability (semiconductor); Medicine; Reliability engineering; Computer science; Intensive care medicine; Economics; Biology; Engineering; Genetics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002847515,0.0007370527,0.000788731,0.001619765,0.0007939779,0.002232476,0.001767088,0.001533219,0.005144358],"category_scores_gemma":[0.0347282,0.0004488146,0.0005146705,0.0005983337,0.0008671618,0.002100331,0.002798,0.001662396,0.001449452],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009823453,"about_ca_system_score_gemma":0.003201426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005750874,"about_ca_topic_score_gemma":0.006386163,"domain_scores_codex":[0.9973096,0.0009109623,0.000159663,0.0004694655,0.0009030929,0.0002472393],"domain_scores_gemma":[0.9782829,0.01419128,0.001102946,0.001695567,0.004043204,0.0006840032],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003918454,0.0008114994,0.03690573,0.0008071059,0.0002050882,0.001760127,0.00130426,0.1525249,0.05215615,0.03082705,0.0161353,0.7026444],"study_design_scores_gemma":[0.0001232712,0.00035479,0.003550746,0.0001443194,0.0001027089,0.0009525066,0.0003009046,0.9256939,0.02733192,0.03384665,0.007538418,0.00005977942],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1876104,0.002361417,0.7760612,0.007110357,0.000728355,0.000355941,0.000776527,0.005048876,0.01994694],"genre_scores_gemma":[0.8887749,0.0002498633,0.107573,0.0007989481,0.0001649498,0.00006074909,0.0003196562,0.0001203882,0.001937415],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005750874,"threshold_uncertainty_score":0.01720959,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1265204214225965,"score_gpt":0.4847186370965837,"score_spread":0.3581982156739871,"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."}}