{"id":"W4412720837","doi":"10.1371/journal.pone.0328061","title":"Differentiation of COVID-19 from other types of viral pneumonia and severity scoring on baseline chest radiographs: Comparison of deep learning with multi-reader evaluation","year":2025,"lang":"en","type":"article","venue":"PLoS ONE","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Health Sciences Centre; Sunnybrook Hospital; University of Toronto; Sunnybrook Health Science Centre; Concordia University","funders":"Sunnybrook Foundation","keywords":"Medicine; Coronavirus disease 2019 (COVID-19); Viral pneumonia; Pneumonia; Generalizability theory; Radiography; Community-acquired pneumonia; Severity of illness; Internal medicine; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Radiology","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":[0.0003716878,0.0001184936,0.0004820481,0.0002701786,0.0000382166,0.000005590405,0.00004706348,0.0000715625,0.00004697678],"category_scores_gemma":[0.001035713,0.0001060806,0.00003993855,0.0002652783,0.00009356496,0.00004395575,0.00002307736,0.0001530556,5.827208e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009188957,"about_ca_system_score_gemma":0.00009679176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001134115,"about_ca_topic_score_gemma":0.00031845,"domain_scores_codex":[0.9987297,0.0001719418,0.0003382199,0.0002301063,0.0004359399,0.00009406191],"domain_scores_gemma":[0.9987066,0.0005343088,0.000281342,0.0002050084,0.0002183469,0.0000543293],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005040618,0.002343715,0.8631099,0.00105495,0.000459411,3.638311e-7,0.001542286,0.001760177,0.1263371,0.000009143608,0.000008344158,0.002870586],"study_design_scores_gemma":[0.003005671,0.0003864957,0.6655174,0.001659817,0.001279103,1.077837e-7,0.0002200765,0.1904482,0.1373581,0.00002207745,0.00001795094,0.00008504235],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.993295,0.0005762373,0.00445022,0.001149125,0.00001777844,0.0004545747,0.00001427457,0.00002901003,0.00001374514],"genre_scores_gemma":[0.9959241,0.00004381948,0.003539077,0.0003683025,0.00002113669,0.00001932504,0.00005433269,0.00001504025,0.00001490332],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1975925,"threshold_uncertainty_score":0.4325843,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08366539472402514,"score_gpt":0.3425783199888204,"score_spread":0.2589129252647953,"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."}}