{"id":"W4410129348","doi":"10.1177/12034754251336238","title":"Evaluating the Diagnostic and Treatment Capabilities of GPT-4 Vision in Dermatology: A Pilot Study","year":2025,"lang":"en","type":"article","venue":"Journal of Cutaneous Medicine and Surgery","topic":"Cutaneous Melanoma Detection and Management","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Medicine; Wilcoxon signed-rank test; Medical physics; Artificial intelligence; Diagnostic accuracy; Computer science; Radiology; Internal medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002619479,0.0007764583,0.0004231427,0.0008754022,0.0001809584,0.000903146,0.001420008,0.001286857,0.002893088],"category_scores_gemma":[0.008563694,0.0002733149,0.0007932006,0.0003590935,0.0003986637,0.0007185185,0.0009171448,0.0008496253,0.0007446997],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001239762,"about_ca_system_score_gemma":0.0008065354,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00499986,"about_ca_topic_score_gemma":0.005213644,"domain_scores_codex":[0.998988,0.000384531,0.00005618647,0.0002591746,0.0002238114,0.00008832412],"domain_scores_gemma":[0.9969976,0.002046476,0.0001306288,0.0003002411,0.0003615874,0.0001635995],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.003529104,0.002765491,0.09585498,0.001084461,0.0005908692,0.0009380761,0.0006117897,0.2312181,0.02667535,0.0008772375,0.009256969,0.6265976],"study_design_scores_gemma":[0.00027463,0.003286379,0.04450336,0.0001345399,0.0002207156,0.001626446,0.0002969651,0.9233603,0.02048801,0.001525827,0.004213943,0.00006890493],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9510074,0.001061985,0.03833098,0.0004450093,0.0001130508,0.0006338794,0.001200821,0.002503343,0.004703465],"genre_scores_gemma":[0.9604386,0.0002221128,0.03656767,0.0001696961,0.000025214,0.0001106861,0.001603999,0.0000525734,0.0008094508],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00499986,"threshold_uncertainty_score":0.01385325,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05507283681032029,"score_gpt":0.3568284464436835,"score_spread":0.3017556096333632,"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."}}