Detecting New Lesions Using a Large Language Model: Applications in Real‐World Multiple Sclerosis Datasets DOI Creative Commons
Shane Poole, Nikki Sisodia, Kanishka Koshal

и другие.

Annals of Neurology, Год журнала: 2025, Номер unknown

Опубликована: Апрель 25, 2025

Objective Neuroimaging is routinely utilized to identify new inflammatory activity in multiple sclerosis (MS). A large language model classify narrative magnetic resonance imaging reports the electronic health record (EHR) as discrete data could provide significant benefits for MS research. The objectives of current study were develop such a prompt and illustrate its research applications through common clinical scenario: monitoring response B‐cell depleting therapy (BCDT). Methods An institutional ecosystem that securely connects healthcare with ChatGPT4 was applied single EHR (2000–2022). (msLesionprompt) developed iteratively refined presence or absence T2‐weighted lesions (newT2w) contrast‐enhancing (CEL). multistep validation included evaluating efficiency (time cost), comparison manually annotated using standard confusion matrix, application identifying predictors newT2w/CEL after BCDT start. Results Accuracy msLesionprompt high detection newT2w (97%) CEL (96.8%). All 14,888 available categorized 4.13 hours ($28); 79% showed no CEL. Data extracted expected suppression by (>97% images an initial “rebaseline” scan). Neighborhood poverty (Area Deprivation Index) identified predictor (newT2w: OR 1.69, 95% CI 1.10–2.59, p = 0.017; CEL: 1.54, 1.01–2.34, 0.046). Interpretation Extracting information from feasible efficient. This approach augment many real‐world analyses disease evolution treatment response. ANN NEUROL 2025

Язык: Английский

Detecting New Lesions Using a Large Language Model: Applications in Real‐World Multiple Sclerosis Datasets DOI Creative Commons
Shane Poole, Nikki Sisodia, Kanishka Koshal

и другие.

Annals of Neurology, Год журнала: 2025, Номер unknown

Опубликована: Апрель 25, 2025

Objective Neuroimaging is routinely utilized to identify new inflammatory activity in multiple sclerosis (MS). A large language model classify narrative magnetic resonance imaging reports the electronic health record (EHR) as discrete data could provide significant benefits for MS research. The objectives of current study were develop such a prompt and illustrate its research applications through common clinical scenario: monitoring response B‐cell depleting therapy (BCDT). Methods An institutional ecosystem that securely connects healthcare with ChatGPT4 was applied single EHR (2000–2022). (msLesionprompt) developed iteratively refined presence or absence T2‐weighted lesions (newT2w) contrast‐enhancing (CEL). multistep validation included evaluating efficiency (time cost), comparison manually annotated using standard confusion matrix, application identifying predictors newT2w/CEL after BCDT start. Results Accuracy msLesionprompt high detection newT2w (97%) CEL (96.8%). All 14,888 available categorized 4.13 hours ($28); 79% showed no CEL. Data extracted expected suppression by (>97% images an initial “rebaseline” scan). Neighborhood poverty (Area Deprivation Index) identified predictor (newT2w: OR 1.69, 95% CI 1.10–2.59, p = 0.017; CEL: 1.54, 1.01–2.34, 0.046). Interpretation Extracting information from feasible efficient. This approach augment many real‐world analyses disease evolution treatment response. ANN NEUROL 2025

Язык: Английский

Процитировано

0