{"title":"False Reassurances from False Negatives: AI Stroke Algorithms and Their Anatomical Blind Spots.","authors":"Abhinav Suri","doi":"10.1148/ryai.260969","DOIUrl":"10.1148/ryai.260969","url":null,"abstract":"","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":"8 6","pages":"e260969"},"PeriodicalIF":20.1,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149325240","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Sang-Jin Park, Jinyoung Choi, Seokwon Kim, Seungeon Song, Insu Park, Dougho Park, Taeyeon Kim, Youjin Lee, Donghoon Yang, Jae Man Cho, Joong Won Yang, Mansu Kim, Heum Dai Kwon, Hong-Gyu Baek, Dae-Chul Cho, Injung Kim
Nalan Karunanayake, Javin Schefflein, Binsheng Zhao, Lawrence H Schwartz
{"title":"Automation Bias and Deskilling in Radiology AI: A Confidence-stratified Triage Model for Workflow Integration.","authors":"Nalan Karunanayake, Javin Schefflein, Binsheng Zhao, Lawrence H Schwartz","doi":"10.1148/ryai.260591","DOIUrl":"https://doi.org/10.1148/ryai.260591","url":null,"abstract":"<p><p>Radiology is adopting artificial intelligence (AI) faster than it is studying what AI does to radiologists' perceptual expertise and interpretive skill. The question has partly shifted from whether these tools perform well on benchmark tasks to what happens to clinical judgment when use of these tools becomes routine. Two risks accompany this transition: automation bias, the tendency to follow AI suggestions without adequate scrutiny, and deskilling, the erosion of perceptual and interpretive ability through disuse, typically revealed when AI support is withdrawn. Incorrect AI output degrades accuracy regardless of experience, the benefit of assistance varies unpredictably across readers and tasks, and performance falls after AI withdrawal; neither risk has been studied longitudinally in radiology. A confidence-stratified triage model is proposed in which calibrated AI confidence, not findings alone, routes cases to low-, intermediate-, or high-confidence streams: output withheld until independent reading, shown with uncertainty signals and a checklist, or shown with random blinded audit. Demonstrated local confidence calibration is a precondition for deployment, and stream thresholds and audit rates should be weighted by task and potential clinical harm, not error probability alone. Rather than applying one workflow to every case, the model redistributes radiologist attention to where AI error is most likely and most consequential, operationalizing four principles: keeping the radiologist in the decision loop, making AI uncertainty visible, inserting targeted cognitive friction, and governing the human-AI interaction as a system. A research agenda to determine whether these countermeasures reduce automation bias and skill erosion in practice is also proposed.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e260591"},"PeriodicalIF":20.1,"publicationDate":"2026-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149289103","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Kartik Bose, Pankaj Gupta, Lisa Adams, Rupesh Agrawal, Mariam Aboian, Soumen Basu, Søren Brunak, Ken Chang, Linda C Chu, Renato Cuocolo, Tugba Akinci D'Antonoli, Salvatore Claudio Fanni, Eduardo Moreno Judice de Mattos Farina, Anthony A Gatti, Satyam Ghodasara, Ricardo A Gonzales, Amit Gupta, Katharina Hoebel, William Hsu, Zekun Jiang, Bardia Khosravi, Michail E Klontzas, Agklinta Kiosia, Neeraj Lalwani, Curtis Langlotz, Hoyeon Lee, Linda Moy, Debnath Pal, Iwan Paolucci, Cody Heath Savage, Krithika Rangarajan, Ronnie Sebro, Abhinav Suri, Ali Tejani, Shuncong Wang, Charles E Kahn
Mohamed Sobhi Jabal, Jikai Zhang, Dominic LaBella, Jessica L Houk, Dylan Zhang, Jeffrey D Rudie, Maciej A Mazurowski, Kirti Magudia, Evan Calabrese
{"title":"Agentic Automation of BT-RADS Scoring Using an End-to-End System for Standardized Glioma Follow-up Assessment.","authors":"Mohamed Sobhi Jabal, Jikai Zhang, Dominic LaBella, Jessica L Houk, Dylan Zhang, Jeffrey D Rudie, Maciej A Mazurowski, Kirti Magudia, Evan Calabrese","doi":"10.1148/ryai.260234","DOIUrl":"https://doi.org/10.1148/ryai.260234","url":null,"abstract":"<p><p>The Brain Tumor Reporting and Data System (BT-RADS) standardizes posttreatment MRI response assessment in patients with diffuse gliomas but requires complex integration of imaging trends, medication effects, and radiation timing. An end-to-end multiagent large language model and convolutional neural network system for automated BTRADS classification was developed and evaluated. Five hundred nine consecutive posttreatment glioma MRI examinations (March 2017-July 2021) from a single center were retrospectively evaluated. An extractor agent identified clinical variables from notes using gpt-oss (OpenAI, 20-billion parameters). A scorer agent applied BTRADS decision logic integrating these with quantitative volumetrics. System accuracy was compared against initial assessments, using expert annotation as the reference standard. Four hundred ninety-two examinations were evaluable. Overall system accuracy was 76.0% (374/492) versus 57.5% (283/492) for initial assessments, an absolute difference of +18.5 percentage points (<i>P</i> < .001). Cohen kappa between the system and the reference standard was 0.71. The system achieved high sensitivity in context-dependent categories (BT-1b; 100%; BT-1a, 92.7%; BT-3a, 87.5%) and moderate sensitivity for threshold-dependent categories (BT-3c, 74.8%; BT-2, 69.2%; BT-4, 69.3%; BT-3b, 57.1%). The BT-4 positive predictive value was 92.9%. These results support further evaluation of this system as a clinical decision-support tool, with local open-weight inference preserving data privacy.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e260234"},"PeriodicalIF":20.1,"publicationDate":"2026-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149239697","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}