{"id":"W4412521411","doi":"10.1016/j.jid.2025.06.1589","title":"The Melanoma Toolkit for Early Detection Improves Accuracy but Fails to Improve the Number Needed to Biopsy","year":2025,"lang":"en","type":"article","venue":"Journal of Investigative Dermatology","topic":"Cell Image Analysis Techniques","field":"Biochemistry, Genetics and Molecular Biology","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"Nature Conservancy of Canada; Oregon Health and Science University","keywords":"Biopsy; Melanoma; Computer science; Medicine; Melanoma diagnosis; Radiology; Cancer research","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008842368,0.0009707938,0.001441559,0.002921409,0.0007271192,0.003107466,0.001636062,0.001413139,0.02038059],"category_scores_gemma":[0.02985751,0.0008117205,0.001555708,0.0009062414,0.0005572596,0.003144403,0.002079476,0.002259505,0.01535111],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00120138,"about_ca_system_score_gemma":0.001680352,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005628928,"about_ca_topic_score_gemma":0.01516581,"domain_scores_codex":[0.9936696,0.002266699,0.0004185538,0.0007156275,0.002483735,0.0004459182],"domain_scores_gemma":[0.9698501,0.01572926,0.001953826,0.00468475,0.006312981,0.001469202],"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":[0.001133135,0.0003555123,0.1352076,0.001170801,0.0003901804,0.0004065729,0.0002954009,0.002730349,0.01666485,0.002578648,0.1024595,0.7366074],"study_design_scores_gemma":[0.0004047159,0.001949084,0.4593161,0.003673337,0.001919354,0.02144503,0.001643742,0.1028623,0.0771883,0.02221377,0.3068844,0.000499797],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.337569,0.04726365,0.3387525,0.04814056,0.00633801,0.001734489,0.01657631,0.05554033,0.1480852],"genre_scores_gemma":[0.647464,0.008080951,0.2986466,0.008349105,0.0007904954,0.0007646787,0.005373369,0.00321828,0.02731258],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02038059,"threshold_uncertainty_score":0.06817985,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.007725779810696973,"score_gpt":0.2828913787299825,"score_spread":0.2751655989192855,"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."}}