{"id":"W4299286150","doi":"10.1177/08465371221131200","title":"U“AI” Testing: User Interface and Usability Testing of a Chest X-ray AI Tool in a Simulated Real-World Workflow","year":2022,"lang":"en","type":"article","venue":"Canadian Association of Radiologists Journal","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; Trillium Health Centre; University of Toronto","funders":"","keywords":"Workflow; Usability; Software deployment; Triage; Medicine; Automation; Computer science; Artificial intelligence; Medical physics; Human–computer interaction; Software engineering; Medical emergency; Database","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":[0.009373651,0.001087504,0.000518239,0.0008999681,0.0005913707,0.001528584,0.001365274,0.001007522,0.002820053],"category_scores_gemma":[0.04891696,0.0004904035,0.0006669826,0.0004168647,0.0007829343,0.001591785,0.00126318,0.0005011334,0.0007733583],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005977839,"about_ca_system_score_gemma":0.0009870577,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001199125,"about_ca_topic_score_gemma":0.001117703,"domain_scores_codex":[0.993511,0.003840542,0.0007198767,0.0008065705,0.0006856745,0.0004363472],"domain_scores_gemma":[0.9561039,0.03273338,0.002206639,0.002942379,0.004881456,0.001132268],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01683841,0.01770913,0.1669963,0.004930348,0.000479313,0.002866323,0.08136009,0.01008802,0.1637914,0.001177003,0.008270348,0.5254934],"study_design_scores_gemma":[0.004410628,0.1561242,0.4710078,0.001517105,0.001148593,0.006138021,0.03022023,0.1204892,0.1668715,0.002161506,0.03874559,0.001165611],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9846888,0.00007481271,0.01202544,0.00008614791,0.00002951551,0.001209366,0.0001235917,0.0007881701,0.0009742999],"genre_scores_gemma":[0.9462409,0.00007552771,0.05029586,0.0001973953,0.0000338445,0.001950615,0.0002513819,0.0001314616,0.0008229597],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009373651,"threshold_uncertainty_score":0.04957318,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1068314061063364,"score_gpt":0.3797614144259395,"score_spread":0.2729300083196031,"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."}}