{"id":"W4283804853","doi":"10.3389/fradi.2022.991683","title":"Improving disease classification performance and explainability of deep learning models in radiology with heatmap generators","year":2022,"lang":"en","type":"article","venue":"Frontiers in Radiology","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; Hospital for Sick Children; University of Toronto","funders":"Hospital for Sick Children; University of Toronto","keywords":"Receiver operating characteristic; Baseline (sea); Artificial intelligence; Medicine; Deep learning; Radiography; Machine learning; Radiology; 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":[0.0006489024,0.0001433455,0.0004843888,0.0003969787,0.0001043401,0.000003101131,0.0001012169,0.00008220782,0.00001073515],"category_scores_gemma":[0.0002298736,0.0001399265,0.0000314817,0.0003242448,0.0002929788,0.00009307641,0.00007410959,0.0004675189,1.195921e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006706288,"about_ca_system_score_gemma":0.0002232881,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001088152,"about_ca_topic_score_gemma":0.00001869221,"domain_scores_codex":[0.9983059,0.000459869,0.0003465299,0.0004695726,0.0001158084,0.0003023188],"domain_scores_gemma":[0.9993109,0.0001397087,0.0001378083,0.000278162,0.0000354304,0.00009796847],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001019797,0.0001159824,0.8907948,0.0001722492,0.00001847536,0.00003914985,0.001115759,0.08841185,0.000381683,0.00009703566,0.0002191687,0.01761406],"study_design_scores_gemma":[0.001837258,0.0006056199,0.3944601,0.00002317563,0.00003721592,0.00006597112,0.001051299,0.6010078,0.00004509239,0.000223726,0.0005076825,0.00013505],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9842513,0.002880876,0.01041117,0.001745718,0.0002320504,0.0004336296,0.000003303325,0.00002742873,0.00001449279],"genre_scores_gemma":[0.9939804,0.0002756202,0.004801439,0.0005962891,0.00003526111,0.0002485131,0.00003040419,0.00001987237,0.00001221072],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5125959,"threshold_uncertainty_score":0.5706037,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01762159431503606,"score_gpt":0.2475747633943266,"score_spread":0.2299531690792906,"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."}}