{"id":"W4391252159","doi":"10.1101/2024.01.24.24301743","title":"Evaluating the Diagnostic and Treatment Recommendation Capabilities of GPT-4 Vision in Dermatology","year":2024,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Dermatology; 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":[0.002231276,0.0008150634,0.0006041517,0.001671592,0.0002320957,0.001335538,0.001299874,0.001894191,0.002390721],"category_scores_gemma":[0.009347245,0.0002828149,0.0008251848,0.0006558998,0.0002663273,0.0008764176,0.0009219926,0.0007920015,0.000820652],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001553615,"about_ca_system_score_gemma":0.000970049,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0128295,"about_ca_topic_score_gemma":0.01112642,"domain_scores_codex":[0.9988676,0.0003576846,0.000060504,0.0003392138,0.0002863375,0.00008861441],"domain_scores_gemma":[0.9975821,0.001493825,0.0001334809,0.0002428389,0.0004037124,0.0001440526],"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":[0.001907924,0.001151394,0.07411391,0.0008875523,0.0005771799,0.000650532,0.0002541526,0.2594386,0.01123186,0.001267232,0.02354972,0.6249698],"study_design_scores_gemma":[0.0000810975,0.000666269,0.01418078,0.0001031318,0.0001034256,0.0004989718,0.0001060414,0.9745995,0.005454553,0.00118328,0.002988625,0.00003433167],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8527126,0.004790405,0.1090842,0.002274584,0.0004086524,0.0008790286,0.004926236,0.01071246,0.01421191],"genre_scores_gemma":[0.9316894,0.0005006886,0.06311677,0.0003858891,0.0000471004,0.0001060394,0.003000652,0.00006241824,0.001091136],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0128295,"threshold_uncertainty_score":0.0255096,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2362795634093349,"score_gpt":0.506050948916747,"score_spread":0.2697713855074121,"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."}}