From the Department of Neurology (J.J., W.G., M.B.F., M.T., K.N., F.A.N., Z.F., S.N., S.S.C., D.B.H., A.J.C., E.S.R., S.F.Z., M.B.W.), Massachusetts General Hospital, Harvard Medical School, Boston; Massachusetts General Hospital Clinical Data Animation Center (CDAC) (J.J., W.G., M.B.F., M.T., F.A.N., Z.F., S.N., S.S.C., D.B.H., S.F.Z., M.B.W.), Boston; National Institute of Health Data Science (S.H.), Peking University, Beijing, China; College of Computing (Z.L., C.Y., J.S.), University of Illinois at Urbana-Champaign; College of Computing (S.A.), Georgia Institute of Technology, Atlanta; Department of Neurology (A.F.S.), University of Wisconsin-Madison; William S. Middleton Memorial Veterans Hospital (A.F.S.), Madison, WI; Yale New Haven Hospital (A.H., J.A.K., E.J.G.), Yale University, CT; Emory University School of Medicine (I.K., A.A.R.R.), Atlanta, GA; Medical University of South Carolina (J.J.H., S.S.), Charleston; University of Manitoba (M.C.N.), Winnipeg, Canada; Johns Hopkins School of Medicine (E.L.J., P.W.K., M.C.C.), Baltimore, MD; University of Arizona College of Medicine (B.L.A.), Phoenix; Brigham and Women's Hospital (R.A.S., J.W.L.), Boston, MA; Mayo Clinic (G.O.), Rochester, MN; Warren Alpert School of Medicine (M.B.D.), Brown University, Providence, RI; University of Nebraska Medical Center (L.A.J., O.T.), Omaha; West Virginia University Hospitals (Z.S.), Morgantown; University of Chicago (H.A.H.), IL; Atrium Health (C.B.S.), Charlotte, NC; Hôpital Erasme (N.G.), Université Libre de Bruxelles, Belgium; Icahn School of Medicine (J.Y.Y.), Mount Sinai, NY; NYU Grossman School of Medicine (M.G.H.), New York; Barrow Neurological Institute (S.T.H.), Phoenix, AZ; Mater Misericordiae University Hospital (J.A.W.), Dublin, Ireland; University of Pennsylvania (J.P.), Philadelphia; and Beth Israel Deaconess Medical Center (M.M.S.), Harvard Medical School, Boston, MA.
Neurology. 2023 Apr 25;100(17):e1750-e1762. doi: 10.1212/WNL.0000000000207127. Epub 2023 Mar 6.
Seizures (SZs) and other SZ-like patterns of brain activity can harm the brain and contribute to in-hospital death, particularly when prolonged. However, experts qualified to interpret EEG data are scarce. Prior attempts to automate this task have been limited by small or inadequately labeled samples and have not convincingly demonstrated generalizable expert-level performance. There exists a critical unmet need for an automated method to classify SZs and other SZ-like events with expert-level reliability. This study was conducted to develop and validate a computer algorithm that matches the reliability and accuracy of experts in identifying SZs and SZ-like events, known as "ictal-interictal-injury continuum" (IIIC) patterns on EEG, including SZs, lateralized and generalized periodic discharges (LPD, GPD), and lateralized and generalized rhythmic delta activity (LRDA, GRDA), and in differentiating these patterns from non-IIIC patterns.
We used 6,095 scalp EEGs from 2,711 patients with and without IIIC events to train a deep neural network, , to perform IIIC event classification. Independent training and test data sets were generated from 50,697 EEG segments, independently annotated by 20 fellowship-trained neurophysiologists. We assessed whether performs at or above the sensitivity, specificity, precision, and calibration of fellowship-trained neurophysiologists for identifying IIIC events. Statistical performance was assessed by the calibration index and by the percentage of experts whose operating points were below the model's receiver operating characteristic curves (ROCs) and precision recall curves (PRCs) for the 6 pattern classes.
matches or exceeds most experts in classifying IIIC events based on both calibration and discrimination metrics. For SZ, LPD, GPD, LRDA, GRDA, and "other" classes, exceeds the following percentages of 20 experts-ROC: 45%, 20%, 50%, 75%, 55%, and 40%; PRC: 50%, 35%, 50%, 90%, 70%, and 45%; and calibration: 95%, 100%, 95%, 100%, 100%, and 80%, respectively.
is the first algorithm to match expert performance in detecting SZs and other SZ-like events in a representative sample of EEGs. With further development, may thus be a valuable tool for an expedited review of EEGs.
This study provides Class II evidence that among patients with epilepsy or critical illness undergoing EEG monitoring, can differentiate (IIIC) patterns from non-IIIC events and expert neurophysiologists.
发作（SZs）和其他类似发作的脑活动模式会损害大脑，并导致住院期间死亡，尤其是当发作持续时间较长时。然而，有资格解读脑电图数据的专家却很稀缺。之前尝试自动执行此任务的方法受到样本量小或标记不足的限制，并且没有令人信服地证明具有可推广的专家级性能。因此，迫切需要一种能够以专家级可靠性对 SZs 和其他类似发作的事件进行分类的自动化方法。本研究旨在开发和验证一种计算机算法，该算法能够匹配专家识别 SZs 和类似发作事件（即脑电图上的“发作-发作间期-损伤连续体”（IIIC）模式）的可靠性和准确性，包括 SZs、局灶性和全身性周期性放电（LPD、GPD）以及局灶性和全身性节律性 delta 活动（LRDA、GRDA），并将这些模式与非-IIIC 模式区分开来。
我们使用来自 2711 名伴有和不伴有 IIIC 事件的患者的 6095 个头皮脑电图数据来训练深度神经网络 ，以执行 IIIC 事件分类。独立的训练和测试数据集是从 50697 个脑电图段生成的，由 20 名接受过神经生理专业培训的医生独立注释。我们评估了 是否达到或超过了接受过神经生理专业培训的专家在识别 IIIC 事件方面的敏感性、特异性、准确性和校准水平。通过校准指数和模型的接收者操作特征曲线（ROC）和精度召回曲线（PRC）下专家比例来评估统计性能，ROC 和 PRC 是针对 6 种模式类别的。
在基于校准和判别指标的情况下， 可以与大多数专家媲美或超过专家的分类性能。对于 SZ、LPD、GPD、LRDA、GRDA 和“其他”类别， 超过以下 20 位专家的 ROC 百分比：45%、20%、50%、75%、55%和 40%；PRC：50%、35%、50%、90%、70%和 45%；校准：95%、100%、95%、100%、100%和 80%。
是第一个在具有代表性的脑电图样本中匹配专家检测 SZs 和其他类似发作事件性能的算法。随着进一步的开发， 可能成为加快脑电图审查的有价值工具。
本研究提供了 II 级证据，表明在接受脑电图监测的癫痫或危重病患者中， 可以区分（IIIC）模式与非-IIIC 事件和专家神经生理学家。
