{"id":"W4399799381","doi":"10.1101/2024.06.19.24309061","title":"An independent, multi-country head-to-head accuracy comparison of automated chest x-ray algorithms for the triage of pulmonary tuberculosis","year":2024,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Centre Hospitalier de l’Université de Montréal","funders":"National Institutes of Health; European and Developing Countries Clinical Trials Partnership; National Institute of Allergy and Infectious Diseases; European Commission","keywords":"Triage; Head (geology); Computer science; Pulmonary tuberculosis; Tuberculosis; Medicine; Algorithm; Radiology; Artificial intelligence; Medical emergency; Pathology; Geology","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":[0.02002823,0.00114658,0.0007905234,0.002566375,0.0006250062,0.001415187,0.001323273,0.001277209,0.001674264],"category_scores_gemma":[0.03685876,0.0006180138,0.001893328,0.001014654,0.001016191,0.001262395,0.002039664,0.0008396152,0.0009544406],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008766326,"about_ca_system_score_gemma":0.0006065233,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002423491,"about_ca_topic_score_gemma":0.003567765,"domain_scores_codex":[0.9855941,0.00846338,0.00121639,0.00221904,0.002112632,0.0003944777],"domain_scores_gemma":[0.956044,0.02203533,0.004652028,0.004725269,0.0113832,0.001160156],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0204443,0.001778144,0.908195,0.000404192,0.003562455,0.00009569058,0.001336811,0.00640046,0.004597235,0.0002534611,0.001517827,0.05141431],"study_design_scores_gemma":[0.001545647,0.015362,0.9422756,0.0001371712,0.00193093,0.0007056075,0.0008776597,0.0257224,0.009057377,0.0002314698,0.001981849,0.000172262],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9922375,0.0004784897,0.004525012,0.00004535166,0.00005329024,0.0004810902,0.001018022,0.0001127812,0.001048415],"genre_scores_gemma":[0.9915994,0.0001117113,0.006071294,0.00005707944,0.00002796715,0.0002735674,0.001570803,0.00004385298,0.000244272],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02002823,"threshold_uncertainty_score":0.1059207,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1127179901596555,"score_gpt":0.4493965221128523,"score_spread":0.3366785319531967,"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."}}