{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007388966,0.001438031,0.0004934415,0.001620014,0.0003462087,0.0008749315,0.00127181,0.001781576,0.003019056],"category_scores_gemma":[0.02923211,0.0002543321,0.001027941,0.0005577288,0.0004881253,0.001154015,0.001236604,0.001439495,0.0003593998],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001933054,"about_ca_system_score_gemma":0.001277471,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003267653,"about_ca_topic_score_gemma":0.002524238,"domain_scores_codex":[0.9964102,0.001610302,0.0003466048,0.0004356201,0.001044044,0.0001531976],"domain_scores_gemma":[0.9718676,0.02261887,0.001300131,0.001178751,0.002587545,0.000447192],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002991695,0.0006947009,0.0193885,0.0011863,0.0005792739,0.000310933,0.0003630935,0.3339645,0.004623541,0.002199326,0.004429003,0.6292692],"study_design_scores_gemma":[0.0001664628,0.0008406218,0.004330226,0.0001689422,0.000130336,0.0001421294,0.00009178041,0.9851768,0.005464522,0.002227287,0.001234214,0.00002670234],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6529112,0.01245631,0.3144009,0.00260182,0.0005881849,0.001301836,0.002172871,0.005937277,0.007629564],"genre_scores_gemma":[0.8846586,0.001229006,0.1106052,0.0002371545,0.00008524179,0.0002642057,0.001603054,0.00007648899,0.001241094],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007388966,"threshold_uncertainty_score":0.03907704,"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."}}