{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004173948,0.0001788447,0.0002781696,0.0001053784,0.0002636718,0.0001359233,0.0005440052,0.0001491445,0.0000022796],"category_scores_gemma":[0.002683182,0.0001077843,0.0002296065,0.0002807629,0.0002572534,0.00002290885,0.0001764805,0.0002041335,0.000007484506],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003679283,"about_ca_system_score_gemma":0.0001654219,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007759174,"about_ca_topic_score_gemma":0.0000819391,"domain_scores_codex":[0.9986502,0.0001857553,0.0005506951,0.0002241981,0.0001137355,0.0002754251],"domain_scores_gemma":[0.9981267,0.0002929557,0.0003892457,0.0003761424,0.0006966924,0.0001182581],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.000178725,0.00001500511,0.0001680364,0.000008878522,0.0002397861,0.000004648462,0.0001465095,0.000002378556,0.978696,0.0001830013,0.01558126,0.004775702],"study_design_scores_gemma":[0.0002914395,0.0002505925,0.00148893,0.00001407301,0.00008571583,0.0001697524,0.0007542723,0.00002643836,0.9605678,0.002302289,0.0339337,0.0001149627],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9583385,0.0001454036,0.02668479,0.01385939,0.000151899,0.0005024061,0.0000043278,0.000009189287,0.0003041069],"genre_scores_gemma":[0.9907076,0.00002839826,0.003323307,0.005114709,0.0001521361,0.000112663,0.000002938766,0.00001571072,0.0005425377],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03236911,"threshold_uncertainty_score":0.4395317,"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."}}