{"id":"W4407573615","doi":"10.1101/2025.02.11.25322087","title":"Evaluating Loon Lens Pro™, an AI-Driven Tool for Full-Text Screening in Systematic Reviews: A Validation Study","year":2025,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Lens (geology); Systematic review; Through-the-lens metering; Data science; Psychology; Computer science; MEDLINE; Political science; Physics; Optics; Law","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":["metaresearch"],"category_scores_codex":[0.4397434,0.002694243,0.005620908,0.01297461,0.00179614,0.006387934,0.004604283,0.002582775,0.004996269],"category_scores_gemma":[0.7169885,0.002342039,0.0098604,0.008604127,0.002394239,0.005388612,0.005654193,0.001838975,0.001329454],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006325204,"about_ca_system_score_gemma":0.02118803,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002926374,"about_ca_topic_score_gemma":0.00865451,"domain_scores_codex":[0.5379017,0.3273093,0.08156346,0.01273741,0.03897908,0.001509113],"domain_scores_gemma":[0.1379863,0.7043129,0.05795951,0.02760286,0.07077501,0.001363454],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.02270096,0.0009839196,0.05657942,0.4010288,0.06823091,0.0007544571,0.01330576,0.01166538,0.00657231,0.006571392,0.01964066,0.3919659],"study_design_scores_gemma":[0.0463781,0.0294522,0.1392906,0.2168072,0.2021201,0.004639096,0.004616455,0.155711,0.03142056,0.02217235,0.1434906,0.003901694],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.324052,0.1094057,0.3506525,0.007678898,0.001743626,0.1614477,0.02125202,0.008916301,0.01485124],"genre_scores_gemma":[0.4378825,0.008355444,0.4474128,0.002250417,0.0003798011,0.09544638,0.006439889,0.0007971959,0.001035581],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5602566,"threshold_uncertainty_score":0.6908962,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5215250717807042,"score_gpt":0.5518716956169177,"score_spread":0.03034662383621356,"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."}}