Otolaryngology--head and neck surgery : official journal of American Academy of Otolaryngology-Head and Neck Surgery · 2025
Automated Machine Learning Differentiation of Pituitary Macroadenomas and Parasellar Meningiomas Using Preoperative Magnetic Resonance Imaging.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Sina Elliott M, Limage Kelsey, Anisman Emma, Pudik Nickolas, Tam Emma, Kahn Chase, Daggumati Srihari, Evans James J, Rabinowitz Mindy R, Rosen Marc R, Nyquist Gurston
Аннотация
INTRODUCTION: Automated machine learning (AutoML) is an artificial intelligence tool that facilitates image recognition model development. This study evaluates the diagnostic performance of AutoML in differentiating pituitary macroadenomas (PA) and parasellar meningiomas (PSM) using preoperative MRI.
STUDY DESIGN: Model development and retrospective analysis.
SETTING: Single academic institution with external validation from a public dataset.
METHODS: 1628 contrast-enhanced T1-weighted MRI sequences from 116 patients (997 PA, 631 PSM) were uploaded to Google Cloud VertexAI AutoML. A single-label classification model was developed using an 80%-10%-10% training-validation-testing split. External validation included 930 PA and 29 PSM images. A subanalysis evaluated the classification of anatomical PSM subtypes (planum sphenoidale [PS] versus tuberculum sellae [TS]). Performance metrics were calculated at 0.25, 0.5, and 0.75 confidence thresholds.
RESULTS: At a 0.5 confidence threshold, the AutoML model achieved an aggregate AUPRC of 0.997, with F1 score, sensitivity, specificity, PPV, and NPV equilibrated to 97.55%. The model achieved strong performance in classifying PA (F1 = 97.98%; sensitivity = 97.00%; specificity = 98.96%) and PSM (F1 = 96.88%; sensitivity = 98.41%; specificity = 95.53%). External validation demonstrated high accuracy (AUPRC = 0.999 for PA; 1.000 for PSM). The PSM subanalysis yielded an aggregate F1 score of 97.30%, with PS and TS classified at 97.44% and 97.14%, respectively.
CONCLUSION: Our customized AutoML model accurately differentiates PAs from PSMs using preoperative MRIs and outperforms traditional ML. It is the first AutoML model specifically trained for parasellar tumor classification. Its highly automated, user-friendly design may facilitate scalable integration into clinical practice.
STUDY DESIGN: Model development and retrospective analysis.
SETTING: Single academic institution with external validation from a public dataset.
METHODS: 1628 contrast-enhanced T1-weighted MRI sequences from 116 patients (997 PA, 631 PSM) were uploaded to Google Cloud VertexAI AutoML. A single-label classification model was developed using an 80%-10%-10% training-validation-testing split. External validation included 930 PA and 29 PSM images. A subanalysis evaluated the classification of anatomical PSM subtypes (planum sphenoidale [PS] versus tuberculum sellae [TS]). Performance metrics were calculated at 0.25, 0.5, and 0.75 confidence thresholds.
RESULTS: At a 0.5 confidence threshold, the AutoML model achieved an aggregate AUPRC of 0.997, with F1 score, sensitivity, specificity, PPV, and NPV equilibrated to 97.55%. The model achieved strong performance in classifying PA (F1 = 97.98%; sensitivity = 97.00%; specificity = 98.96%) and PSM (F1 = 96.88%; sensitivity = 98.41%; specificity = 95.53%). External validation demonstrated high accuracy (AUPRC = 0.999 for PA; 1.000 for PSM). The PSM subanalysis yielded an aggregate F1 score of 97.30%, with PS and TS classified at 97.44% and 97.14%, respectively.
CONCLUSION: Our customized AutoML model accurately differentiates PAs from PSMs using preoperative MRIs and outperforms traditional ML. It is the first AutoML model specifically trained for parasellar tumor classification. Its highly automated, user-friendly design may facilitate scalable integration into clinical practice.