{"id":"W4414442425","doi":"10.1007/978-3-032-05825-6_3","title":"What Can We Learn from Inter-Annotator Variability in Skin Lesion Segmentation?","year":2025,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Cutaneous Melanoma Detection and Management","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Segmentation; Skin lesion; Leverage (statistics); Pattern recognition (psychology); Feature (linguistics); Lesion; Image segmentation; Association (psychology)","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02446238,0.001727002,0.002472194,0.002665956,0.0009555161,0.005126773,0.002725253,0.003588653,0.002225883],"category_scores_gemma":[0.08713607,0.001103089,0.002248961,0.003169935,0.001915039,0.01368217,0.002356276,0.00538409,0.003517136],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001199803,"about_ca_system_score_gemma":0.001180345,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004107278,"about_ca_topic_score_gemma":0.0080093,"domain_scores_codex":[0.9898652,0.004693537,0.000518052,0.002934402,0.001601688,0.0003871921],"domain_scores_gemma":[0.9275503,0.05778985,0.00267304,0.005825212,0.005616976,0.0005446141],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0006072405,0.0002018738,0.04112004,0.001242819,0.00159443,0.0002149089,0.001189947,0.02776486,0.00773237,0.003961995,0.03413926,0.8802302],"study_design_scores_gemma":[0.0000770866,0.000465475,0.06934741,0.001288266,0.001501184,0.001084793,0.002209476,0.6623813,0.01949946,0.2007053,0.04107634,0.0003639089],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1459176,0.04800998,0.7558872,0.02070554,0.003308799,0.000195724,0.00659586,0.006587392,0.01279195],"genre_scores_gemma":[0.6919327,0.01463705,0.2592627,0.004084388,0.003183261,0.0003385397,0.01293352,0.003094118,0.0105337],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9755376,"threshold_uncertainty_score":0.129371,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01480828998132474,"score_gpt":0.2680472393843398,"score_spread":0.2532389494030151,"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."}}