{"id":"W3207181646","doi":"10.2196/33049","title":"Differential Biases and Variabilities of Deep Learning–Based Artificial Intelligence and Human Experts in Clinical Diagnosis: Retrospective Cohort and Survey Study","year":2021,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Yonsei University College of Medicine; Ministry of Science and ICT, South Korea; National Research Foundation of Korea; Ministry of Science, ICT and Future Planning; Inje University; Yonsei University; National Research Foundation","keywords":"Artificial intelligence; Medicine; Machine learning; Population; Test (biology); Medical diagnosis; Medical physics; Computer science; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003593942,0.0002769917,0.0004310965,0.001692415,0.0004774846,0.000989044,0.0005362948,0.0007345501,0.001469391],"category_scores_gemma":[0.01531753,0.0003993435,0.0006636666,0.001470604,0.0006784303,0.001458926,0.0009000006,0.0008073182,0.0004549933],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005939473,"about_ca_system_score_gemma":0.0003730292,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00311167,"about_ca_topic_score_gemma":0.002535723,"domain_scores_codex":[0.9968078,0.0009354193,0.0004320485,0.0008146565,0.0006123147,0.0003977462],"domain_scores_gemma":[0.9907203,0.00335505,0.002813397,0.001002185,0.001488782,0.0006202172],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004752941,0.00002535482,0.9988648,0.000006173438,0.00002964132,0.00003124218,0.0001343631,0.0000248629,0.00002988748,0.00001390576,0.00006120915,0.0007310923],"study_design_scores_gemma":[0.000007369034,0.0001746123,0.9972478,0.00001567675,0.00004563843,0.0003734039,0.0009628261,0.0006479832,0.00008292844,0.00007115977,0.0003586804,0.00001199553],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9991543,0.0001949622,0.0001774087,0.00003401121,0.000004631481,0.00001353712,0.0001922587,0.000002277714,0.0002265612],"genre_scores_gemma":[0.999525,0.00007992515,0.00007221607,0.00003146704,0.000008107372,0.00001591734,0.0001844434,0.000002608173,0.00008042812],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003593942,"threshold_uncertainty_score":0.01900679,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08705776469855132,"score_gpt":0.410996291556137,"score_spread":0.3239385268575857,"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."}}