{"id":"W4317214916","doi":"10.2196/38412","title":"Agreement Between Experts and an Untrained Crowd for Identifying Dermoscopic Features Using a Gamified App: Reader Feasibility Study","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Cutaneous Melanoma Detection and Management","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Cancer Institute","keywords":"Crowdsourcing; Inter-rater reliability; Bottleneck; Reliability (semiconductor); Identification (biology); Computer science; Artificial intelligence; Gold standard (test); Similarity (geometry); Pattern recognition (psychology); Medicine; Image (mathematics); Mathematics; Statistics; Radiology","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":[0.0007107102,0.0001869422,0.0003769217,0.0002091789,0.0001943727,0.00008374796,0.0001139591,0.0001484384,0.00008130392],"category_scores_gemma":[0.0001768871,0.0001522484,0.00006243982,0.0002775586,0.00008497221,0.0001535823,0.0001089682,0.0001903639,0.000009985853],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001065354,"about_ca_system_score_gemma":0.00008152987,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002760396,"about_ca_topic_score_gemma":0.00005359935,"domain_scores_codex":[0.9980234,0.00004592267,0.0006237922,0.0001947744,0.0007566863,0.0003554068],"domain_scores_gemma":[0.9989182,0.0001119983,0.0001229114,0.0003498118,0.00007551058,0.0004215621],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.002490112,0.005420685,0.1755607,0.01631056,0.003064068,0.001231337,0.3947751,0.00005349457,0.00361741,0.0005471322,0.05003216,0.3468973],"study_design_scores_gemma":[0.02810568,0.006184091,0.4864693,0.001279925,0.00126849,0.0005906129,0.3828837,0.07706577,0.0009586372,0.0006151447,0.01314516,0.00143345],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9916865,0.00001914342,0.004191644,0.0003183734,0.0002192102,0.002936162,0.00000741473,0.0002620386,0.0003595784],"genre_scores_gemma":[0.9967193,0.000008250284,0.001468779,0.0007284677,0.0002825808,0.0001638641,0.00007601141,0.00001996378,0.0005327953],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3454638,"threshold_uncertainty_score":0.6208511,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1095782061467511,"score_gpt":0.412816470107709,"score_spread":0.3032382639609579,"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."}}