{"id":"W3207796226","doi":"10.1007/s10140-021-01954-x","title":"Limited generalizability of deep learning algorithm for pediatric pneumonia classification on external data","year":2021,"lang":"en","type":"article","venue":"Emergency Radiology","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Generalizability theory; Medicine; Artificial intelligence; Test set; Convolutional neural network; Radiography; Receiver operating characteristic; Pneumonia; Deep learning; Data set; Set (abstract data type); Test (biology); Machine learning; Pattern recognition (psychology); Radiology; Computer science; Statistics; Internal medicine; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"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":[0.0005380454,0.0001528851,0.0003941632,0.0001379566,0.00008266648,0.000003243151,0.0002325971,0.0001630889,0.000412953],"category_scores_gemma":[0.00188855,0.0001533332,0.0001178045,0.0003651804,0.00004835484,0.00005195006,0.00009346704,0.0002283304,0.00001514026],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008920749,"about_ca_system_score_gemma":0.0001613827,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003498639,"about_ca_topic_score_gemma":0.00001581622,"domain_scores_codex":[0.9980679,0.0002849237,0.0005500272,0.0006574686,0.0001672623,0.000272419],"domain_scores_gemma":[0.9981468,0.0003810113,0.0002197065,0.0009343873,0.0002211034,0.0000970202],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002394984,0.001558098,0.5673342,0.0005471846,0.000244696,0.00006169435,0.0004740529,0.0006736204,0.1250916,0.0004420608,0.03256246,0.2707708],"study_design_scores_gemma":[0.001822244,0.0009421429,0.6735043,0.00002741759,0.0004928742,0.00009772601,0.00006115903,0.2656134,0.002654513,0.0003665118,0.05411508,0.0003026455],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.857007,0.004971995,0.1230519,0.01150306,0.002419335,0.0006921589,0.0001223901,0.000134803,0.00009743166],"genre_scores_gemma":[0.9297794,0.007920112,0.0523237,0.002851154,0.003123906,0.0001781952,0.002913889,0.0001052959,0.0008043312],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2704682,"threshold_uncertainty_score":0.6252745,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08301160984302676,"score_gpt":0.3652829491124194,"score_spread":0.2822713392693926,"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."}}