{"id":"W4409351262","doi":"10.1117/1.jmi.12.5.051803","title":"Correlation of objective image quality metrics with radiologists’ diagnostic confidence depends on the clinical task performed","year":2025,"lang":"en","type":"article","venue":"Journal of Medical Imaging","topic":"Radiology practices and education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"Natural Sciences and Engineering Research Council of Canada; Research Nova Scotia; Dalhousie University; Exxon Mobil Corporation","keywords":"Medicine; Correlation; Image quality; Confidence interval; Quality (philosophy); Medical physics; Radiology; Task (project management); Artificial intelligence; Image (mathematics); Internal medicine","routes":{"ca_aff":true,"ca_fund":true,"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"],"consensus_categories":[],"category_scores_codex":[0.007912074,0.00009609894,0.0004410165,0.0001899541,0.00007838476,0.00001679,0.0002090937,0.0001115111,0.0001129216],"category_scores_gemma":[0.07602758,0.00005017139,0.0001234418,0.0003683144,0.0005480859,0.0001720208,0.00002605445,0.001248734,0.000003693275],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001020145,"about_ca_system_score_gemma":0.001019199,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005029314,"about_ca_topic_score_gemma":0.000003117391,"domain_scores_codex":[0.997494,0.0005893303,0.0008821489,0.0001374074,0.000747589,0.0001495527],"domain_scores_gemma":[0.980476,0.01770924,0.0009435058,0.0002049153,0.000511154,0.0001552477],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001201851,0.0004895095,0.9216357,0.00008070936,0.0003294602,0.0001609852,0.0005318504,0.00002668804,0.0001380314,0.002205921,0.01134512,0.06185414],"study_design_scores_gemma":[0.002567186,0.0005619351,0.980077,0.0009785288,0.0005007688,0.000893185,0.003659373,0.008604896,0.0002147153,0.0005844283,0.001276009,0.00008199803],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8686826,0.001107605,0.06163464,0.06489186,0.001015391,0.0002067554,0.000001248688,0.000008162625,0.002451798],"genre_scores_gemma":[0.9956444,0.001004112,0.0006190235,0.002359201,0.0002995624,0.000002845393,0.000002585762,0.000005198053,0.00006307959],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1269619,"threshold_uncertainty_score":0.9317554,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04908256126639332,"score_gpt":0.4387362302467725,"score_spread":0.3896536689803792,"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."}}