{"id":"W4410225779","doi":"10.1371/journal.pone.0322051","title":"The impact of AI feedback on the accuracy of diagnosis, decision switching and trust in radiography","year":2025,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"Workforce; Software deployment; Decision support system; Artificial intelligence; Medicine; Medical education; Computer science; Binary classification; Diagnostic accuracy; Medical physics; Psychology; Radiology; Political science","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":[0.02396404,0.0003359378,0.0004236702,0.0008233811,0.0006458425,0.00238999,0.0006221528,0.0007995303,0.002530447],"category_scores_gemma":[0.3093939,0.0003128925,0.0004012851,0.0003535997,0.001168422,0.001759109,0.001546068,0.001388894,0.0002897865],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001777945,"about_ca_system_score_gemma":0.001627717,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003706525,"about_ca_topic_score_gemma":0.003639035,"domain_scores_codex":[0.9748355,0.01824364,0.001188801,0.0007918366,0.004118464,0.0008218148],"domain_scores_gemma":[0.5168014,0.4344967,0.02147581,0.006787193,0.01466197,0.005776926],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006689687,0.003584551,0.5661965,0.0009023217,0.0003514787,0.000651753,0.05478289,0.005199887,0.007401953,0.0008459428,0.002147773,0.3512452],"study_design_scores_gemma":[0.0003507702,0.01280439,0.8987176,0.0008080316,0.0004944879,0.0009895312,0.03251721,0.03353057,0.01133796,0.003161543,0.004908235,0.0003796691],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9962813,0.0001290413,0.001428186,0.0004807118,0.00001925376,0.00004199885,0.00002610645,0.00005108139,0.001542243],"genre_scores_gemma":[0.9985405,0.00004695564,0.001122184,0.00005437223,0.000007683807,0.00001424604,0.00001435256,0.000006029203,0.0001935737],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02396404,"threshold_uncertainty_score":0.1267355,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.123337058392948,"score_gpt":0.4131830910763942,"score_spread":0.2898460326834463,"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."}}