{"id":"W4410361163","doi":"10.2196/67457","title":"Automating Colon Polyp Classification in Digital Pathology by Evaluation of a “Machine Learning as a Service” AI Model: Algorithm Development and Validation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"AI in cancer detection","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria; University of Alberta","funders":"","keywords":"Preprint; Service (business); Computer science; Service model; Artificial intelligence; Digital pathology; Medicine; World Wide Web","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003801916,0.0007742823,0.0005979736,0.00116001,0.0003590625,0.0009816132,0.0009193124,0.000923554,0.001293837],"category_scores_gemma":[0.009562876,0.0002172513,0.0006045797,0.0006785195,0.0004053434,0.0006921699,0.0007837397,0.00074498,0.0004666076],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002094813,"about_ca_system_score_gemma":0.001847435,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.013132,"about_ca_topic_score_gemma":0.006685688,"domain_scores_codex":[0.998953,0.0003633332,0.00007412607,0.0002090344,0.0003176961,0.000082894],"domain_scores_gemma":[0.9943243,0.00350962,0.0003156392,0.000388858,0.001346697,0.0001149483],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0005446462,0.0006735011,0.02674648,0.0002171065,0.0001679222,0.0001149858,0.0001377915,0.6210243,0.00591231,0.001427402,0.001802625,0.3412309],"study_design_scores_gemma":[0.00001141245,0.00006945233,0.0009327389,0.000005251837,0.000008922939,0.00002086712,0.00001317198,0.997054,0.001544528,0.0001276658,0.0002087211,0.0000033497],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.646354,0.0007100349,0.3427576,0.0004959154,0.00008155699,0.0007025648,0.0003673709,0.004613238,0.003917826],"genre_scores_gemma":[0.8087987,0.0001623752,0.1887375,0.000116335,0.00002109483,0.0002673436,0.0006053683,0.00009522508,0.001196053],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.013132,"threshold_uncertainty_score":0.02611113,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07303212742839817,"score_gpt":0.4331372897543279,"score_spread":0.3601051623259298,"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."}}