{"id":"W3169686285","doi":"10.2147/opth.s312236","title":"Evaluation of Explainable Deep Learning Methods for Ophthalmic Diagnosis","year":2021,"lang":"en","type":"article","venue":"Clinical ophthalmology","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":52,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Compute Canada; Nvidia","keywords":"Medicine; Drusen; Deep learning; Artificial intelligence; Ranibizumab; Optometry; Attribution; Ophthalmology; Machine learning; Retinal; Computer science; Surgery; Psychology","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008292022,0.0001220375,0.0007702189,0.00008913672,0.00006908023,0.00000820325,0.00008786045,0.0001968936,0.001346015],"category_scores_gemma":[0.02724594,0.0001107834,0.0005183553,0.0003006528,0.0001595735,0.00003983887,0.00006243476,0.0003041718,0.0000174734],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004646235,"about_ca_system_score_gemma":0.0002531125,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001490818,"about_ca_topic_score_gemma":2.45003e-7,"domain_scores_codex":[0.996276,0.001993037,0.0007790622,0.0004366347,0.0002645509,0.0002507723],"domain_scores_gemma":[0.99412,0.003571842,0.0002606283,0.0003421083,0.001576608,0.0001288105],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002303765,0.001092322,0.5817014,0.0002091955,0.0008187439,0.0003371018,0.00009740685,0.000455662,0.003053687,0.0001698442,0.0002219528,0.4116123],"study_design_scores_gemma":[0.01821695,0.00925524,0.4901124,0.0007052216,0.02113506,0.01503555,0.003019042,0.3315498,0.05620725,0.03084053,0.02280428,0.001118694],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9793969,0.003925614,0.008276722,0.001264278,0.0003561408,0.0002915767,0.000001834096,0.00002075881,0.006466193],"genre_scores_gemma":[0.9252543,0.0001673258,0.07277718,0.00009917816,0.0001781144,0.0001534856,0.00006100321,0.00002175569,0.001287661],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4104936,"threshold_uncertainty_score":0.9995669,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2424441744608055,"score_gpt":0.5720618935828213,"score_spread":0.3296177191220158,"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."}}