{"id":"W4414444153","doi":"10.1002/ajum.70026","title":"Development of a Novel Machine Learning Model for Automatic Assessment of Quality of Transvaginal Ultrasound Images From Multi‐Annotator Labels","year":2025,"lang":"en","type":"article","venue":"Australasian Journal of Ultrasound in Medicine","topic":"Gynecological conditions and treatments","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"University of Adelaide; Australian Government","keywords":"Transvaginal ultrasound; Quality (philosophy); Quality assessment; Ultrasound","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001240199,0.0002398115,0.001356732,0.0003970613,0.00005292768,0.000004785884,0.0001813358,0.0001198724,0.0002005113],"category_scores_gemma":[0.00179032,0.0001708261,0.0001956503,0.0003500195,0.0003017218,0.0001076285,0.00001110845,0.0004119347,1.51175e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001542154,"about_ca_system_score_gemma":0.0005288633,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006959072,"about_ca_topic_score_gemma":0.00004511024,"domain_scores_codex":[0.9967858,0.0001015397,0.002174876,0.000201931,0.0004908754,0.0002449573],"domain_scores_gemma":[0.9959785,0.002111636,0.001112872,0.0001825815,0.000479314,0.0001351035],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.0002385607,0.003369422,0.356602,0.00104604,0.001212121,0.00001429224,0.002575265,0.001197902,0.6309249,0.0004940425,0.00004980592,0.002275604],"study_design_scores_gemma":[0.01952847,0.001433629,0.9516451,0.002461717,0.0007636333,0.00003020515,0.0026351,0.004142864,0.01662455,0.0005505937,0.00003683494,0.0001473511],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.902319,0.0002598463,0.09611169,0.0004836835,0.00008633698,0.0004336879,0.0001420333,0.000007230528,0.0001564821],"genre_scores_gemma":[0.7728426,0.00004737572,0.2268397,0.00003729948,0.00001252964,0.00001243187,0.00007449739,0.00001059474,0.0001229505],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6143004,"threshold_uncertainty_score":0.6966084,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06695891411644737,"score_gpt":0.4014191997529949,"score_spread":0.3344602856365475,"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."}}