{"id":"W4408692716","doi":"10.2196/67299","title":"Assessing the Diagnostic Accuracy of ChatGPT-4 in Identifying Diverse Skin Lesions Against Squamous and Basal Cell Carcinoma","year":2025,"lang":"en","type":"article","venue":"JMIR Dermatology","topic":"Cutaneous Melanoma Detection and Management","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Basal cell carcinoma; Basal cell; Diagnostic accuracy; Dermatoscopy; Basal (medicine); Pathology; Dermatology; Skin cancer; Medicine; Cancer research; Internal medicine; Cancer; Melanoma","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00008566786,0.0001060957,0.0002526006,0.0002768571,0.0001073649,0.00002735502,0.00008290759,0.00008186557,0.00002513809],"category_scores_gemma":[0.0003063225,0.00008412701,0.00004988374,0.0002439575,0.0001053258,0.00006944892,0.000161661,0.0001906134,0.000008715006],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004062099,"about_ca_system_score_gemma":0.00005772374,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009765288,"about_ca_topic_score_gemma":0.0001166777,"domain_scores_codex":[0.9991421,0.0001052587,0.0002817138,0.0001895424,0.00009276938,0.0001885798],"domain_scores_gemma":[0.9987039,0.0008989854,0.00009615708,0.0002274072,0.00003217816,0.00004136329],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000203103,0.001150241,0.8760304,0.003329294,0.0001030653,0.006448272,0.006097493,0.00006319126,0.01748778,0.001759353,0.01337321,0.07395463],"study_design_scores_gemma":[0.002489751,0.00004938183,0.9713803,0.0001929288,0.0001136208,0.000526534,0.007077251,0.001692214,0.01045683,0.0001748159,0.00570305,0.0001433279],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.991974,0.0004582453,0.000232169,0.002070145,0.0002626246,0.0005724865,0.000001319171,0.00002368599,0.004405302],"genre_scores_gemma":[0.9985299,0.00008539522,0.00005456799,0.0009772622,0.00001721014,0.00005331261,0.000004777584,0.0000080593,0.0002695225],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09534993,"threshold_uncertainty_score":0.34306,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02050781727261161,"score_gpt":0.32019733300373,"score_spread":0.2996895157311185,"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."}}