{"id":"W4401367002","doi":"10.2196/63010","title":"Comparative Study to Evaluate the Accuracy of Differential Diagnosis Lists Generated by Gemini Advanced, Gemini, and Bard for a Case Report Series Analysis: Cross-Sectional Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Series (stratigraphy); Computer science; Data mining; Geology","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004421106,0.000463145,0.000863208,0.005688841,0.0008817544,0.001537357,0.0008049706,0.0007093746,0.003218497],"category_scores_gemma":[0.02543755,0.000475769,0.001114619,0.002871175,0.0007646045,0.002672322,0.001132019,0.0006141394,0.0007212452],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009511173,"about_ca_system_score_gemma":0.0006751035,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001674316,"about_ca_topic_score_gemma":0.002036378,"domain_scores_codex":[0.9942663,0.001137234,0.001538195,0.001165238,0.001480029,0.0004130107],"domain_scores_gemma":[0.9774919,0.008384578,0.007991404,0.00118854,0.003966593,0.0009770626],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001031002,0.00004901288,0.9946226,0.00010041,0.00007728712,0.0007958797,0.0004551503,0.00003635352,0.0001516739,0.00004486476,0.0001929174,0.003370704],"study_design_scores_gemma":[0.00002190754,0.0007598759,0.9708282,0.0001499661,0.0003742104,0.01866882,0.00564357,0.0008841644,0.0005870503,0.0001599599,0.001873895,0.00004844306],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9963728,0.001755182,0.0005455536,0.00006409183,0.00003191512,0.00008679202,0.0003802726,0.00001208493,0.0007512395],"genre_scores_gemma":[0.9978464,0.0005495777,0.0005920788,0.00005152525,0.00004434817,0.00007579509,0.0006465944,0.000009270268,0.0001845401],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9955789,"threshold_uncertainty_score":0.02338129,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1525138604214925,"score_gpt":0.5249425515847759,"score_spread":0.3724286911632834,"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."}}