{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003413517,0.001658464,0.0008541978,0.0009747734,0.000354906,0.001131183,0.0009985608,0.001579087,0.0007693351],"category_scores_gemma":[0.007600509,0.0003634747,0.001159196,0.0005482255,0.0005928975,0.00156514,0.001110429,0.002057935,0.0002535491],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001344229,"about_ca_system_score_gemma":0.0008623091,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01154813,"about_ca_topic_score_gemma":0.007272169,"domain_scores_codex":[0.9992059,0.0002910161,0.00005305876,0.0002379483,0.0001042388,0.000108],"domain_scores_gemma":[0.9966254,0.002297442,0.0002497667,0.0003479977,0.0003676426,0.0001118345],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008948115,0.0004397766,0.02010576,0.0001888163,0.0002878265,0.0002308119,0.0002766546,0.7690009,0.006969491,0.001814628,0.003315826,0.1964747],"study_design_scores_gemma":[0.00001555779,0.000145971,0.001975326,0.00001673102,0.00003155912,0.00002604711,0.00002308071,0.9937724,0.002372236,0.00137359,0.0002317767,0.00001559363],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7330379,0.00314582,0.2542486,0.001758588,0.0003115171,0.0001642612,0.000832769,0.003897819,0.002602685],"genre_scores_gemma":[0.971601,0.0003078112,0.02594759,0.0002425066,0.0000583783,0.00005257253,0.0007678401,0.00005245704,0.0009698387],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01154813,"threshold_uncertainty_score":0.02296185,"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."}}