{"id":"W4390419555","doi":"10.2196/46817","title":"Comparison of the Discrimination Performance of AI Scoring and the Brixia Score in Predicting COVID-19 Severity on Chest X-Ray Imaging: Diagnostic Accuracy Study","year":2023,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; Coronavirus disease 2019 (COVID-19); Receiver operating characteristic; Area under the curve; Pneumonia; Severity of illness; Area under curve; Prospective cohort study; Internal medicine; Disease; Radiology","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":[0.006406194,0.0006618226,0.0006370154,0.002698245,0.0002102428,0.001214636,0.000605953,0.0006648999,0.0008637494],"category_scores_gemma":[0.01769676,0.0002596281,0.0007245402,0.000881615,0.0003880502,0.0006998739,0.0006966554,0.0005204377,0.0003033127],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003371562,"about_ca_system_score_gemma":0.0003022821,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008801571,"about_ca_topic_score_gemma":0.0008995436,"domain_scores_codex":[0.996734,0.001583813,0.0003756067,0.0004785495,0.0006700034,0.0001579832],"domain_scores_gemma":[0.9889842,0.006765976,0.001730727,0.0005893675,0.001369597,0.0005602543],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006414073,0.0001098026,0.9882511,0.0000534342,0.000205724,0.00005989337,0.00008685352,0.0005937337,0.0006684746,0.0000535805,0.0001449124,0.009131012],"study_design_scores_gemma":[0.00006931434,0.0009899389,0.9706669,0.00004023311,0.0002398527,0.0008211829,0.0002145796,0.02486738,0.001409281,0.0001734556,0.0004791775,0.00002884474],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9956904,0.0006381336,0.001992368,0.00005755187,0.00003134824,0.00006925411,0.000274579,0.00004392393,0.001202444],"genre_scores_gemma":[0.997768,0.0001387982,0.001579245,0.00001802474,0.00001820741,0.00002846864,0.000299999,0.000004945099,0.0001444332],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006406194,"threshold_uncertainty_score":0.03387958,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1443200733992593,"score_gpt":0.4913833580452995,"score_spread":0.3470632846460403,"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."}}