{"id":"W4318699054","doi":"10.1136/bjo-2022-322183","title":"Ensemble of deep convolutional neural networks is more accurate and reliable than board-certified ophthalmologists at detecting multiple diseases in retinal fundus photographs","year":2023,"lang":"en","type":"article","venue":"British Journal of Ophthalmology","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Michael's Hospital; Sunnybrook Hospital; University of Toronto; Kensington Health; University Health Network; University of British Columbia","funders":"Canadian Medical Association; Foundation Fighting Blindness","keywords":"Medicine; Convolutional neural network; Fundus (uterus); Optometry; Retinal; Ophthalmology; Artificial intelligence; Certification; Computer vision; Computer science","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002721618,0.0007430444,0.0006236234,0.001112102,0.0003130629,0.0008151679,0.0003941504,0.000871364,0.0008694262],"category_scores_gemma":[0.00563515,0.0002620514,0.0004436375,0.0003814195,0.0001733908,0.0009178157,0.0007192358,0.0005830484,0.0003825858],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005279467,"about_ca_system_score_gemma":0.0005636939,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007235675,"about_ca_topic_score_gemma":0.01457326,"domain_scores_codex":[0.9988625,0.0001926155,0.0000795548,0.0004528547,0.0002706701,0.000141847],"domain_scores_gemma":[0.9967939,0.0008680316,0.0004888973,0.0003759952,0.001201333,0.0002718491],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001033996,0.0004250609,0.6055868,0.0001380031,0.0006704167,0.0003178381,0.0001929194,0.03271558,0.0194534,0.0002386136,0.008484109,0.3307433],"study_design_scores_gemma":[0.000101141,0.000786515,0.4341822,0.0001594271,0.0005394729,0.00135402,0.0003756428,0.527288,0.02845938,0.001025801,0.005633793,0.00009458002],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9569077,0.001251767,0.03538458,0.0004407744,0.0001272697,0.00007333269,0.0005663531,0.001273945,0.003974327],"genre_scores_gemma":[0.9853562,0.000208617,0.01218693,0.0001681501,0.00004581605,0.00001444534,0.0008898669,0.00003317912,0.001096844],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007235675,"threshold_uncertainty_score":0.01439351,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03146420734778214,"score_gpt":0.3036909760647548,"score_spread":0.2722267687169726,"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."}}