{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005874227,0.000209612,0.0008312414,0.0004432267,0.0001544392,0.00003778391,0.000161032,0.0001762661,0.0001871304],"category_scores_gemma":[0.0007862159,0.0002194064,0.0003250763,0.0007802101,0.0004140597,0.0001388396,0.0001798398,0.0005569385,0.00000253894],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006794983,"about_ca_system_score_gemma":0.0000580841,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009120216,"about_ca_topic_score_gemma":0.00001669824,"domain_scores_codex":[0.9976916,0.0001917877,0.0008951813,0.0003521497,0.0003460297,0.0005232057],"domain_scores_gemma":[0.9981995,0.0004335936,0.0005929362,0.0001796389,0.0003414754,0.0002528801],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001228814,0.0001672994,0.9533099,0.0001214272,0.0001709423,0.03571692,0.0001194122,0.002983844,0.003742987,9.266396e-7,0.000410622,0.002026916],"study_design_scores_gemma":[0.001955252,0.0004832485,0.7370908,0.000378234,0.0001869415,0.2010911,0.000424468,0.05741782,0.0006522037,0.0001424046,0.000007360966,0.000170218],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.993677,0.005296933,0.00007023632,0.0003558949,0.0002098495,0.0001502832,0.00002049577,0.00001964685,0.0001997017],"genre_scores_gemma":[0.9987598,0.000380642,0.0003195529,0.00005458313,0.0001336452,0.00000890762,0.00003185655,0.00003057366,0.0002804457],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2162191,"threshold_uncertainty_score":0.8947133,"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."}}