可解释的机器学习模型用于预测抗体阳性自身免疫脑炎患者的预后
Junshuang Guo1, Ruirui Dong2, Ruike Zhang2
1Neuro-Intensive Care Unit of the First Affiliated Hospital of Zhengzhou University, Zhengzhou, Henan Province, China; Department of Immunology, School of Basic Medical Science, Central South University, Changsha City, Hunan Province, China.
Journal of affective disorders
|October 7, 2024
概括
机器学习模型,特别是Random Forest和XGBoost,可以准确预测自身免疫性脑炎的预后. 这些模型为这种罕见的神经疾病提供了良好的性能,临床适用性和可解释性.
科学领域:
- 神经学 神经学
- 人工智能的人工智能
- 医学预测 医学预测
背景情况:
- 自身免疫性脑炎 (AE) 是一种罕见的神经系统疾病,其全球疾病负担正在增加.
- 准确的预后预测对于管理AE患者至关重要.
- 现有的预后模型可能会从先进的机器学习方法中受益.
研究的目的:
- 开发和验证机器学习 (ML) 模型,用于预测抗体阳性自身免疫脑炎 (AE) 患者的预后.
- 为了比较9种不同的ML方法在AE预后预测中的性能.
- 使用可解释的AI技术识别AE预后的关键预测因素.
主要方法:
- 利用了来自全球疾病负担 (GBD) 研究的187名AE患者的数据集.
- 开发了使用九个ML算法的预测模型:决策树 (DT),随机森林 (RF),极端梯度增强 (XGBoost),K-最近邻居 (KNN),支持向量机 (SVM),天真湾 (NB),神经网络 (NN),光梯度增强机 (LGBM) 和后勤回归 (LR).
- 已验证的歧视,校准和临床适用性的模型. 采用沙普利增量解释 (SHAP) 来解释模型的可解释性.
主要成果:
- 脑炎的全球负担,包括死亡,发病率和流行率,从2010年到2021年显示出日益增长的趋势.
- 随机森林 (RF) 获得了最高的精度 (0.860) 和F1得分 (0.844),紧随其后的是XGBoost (精度0.826,F1得分0.807).
- SHAP分析确定了感染,CSF单细胞百分比和前白蛋白作为AE预后的重要预测因素.
结论:
- 射频和XGBoost模型表现出强大的性能,良好的区分,校准和临床适用性,用于预测AE预后.
- 这些ML模型提供了有价值的解释性,有助于临床决策.
- 尽管罕见疾病固有的样本规模很小,但开发的模型对AE患者管理有希望.
相关概念视频
Encephalitis l: Introduction
27
Encephalitis is inflammation of the brain parenchyma, most often due to infections or autoimmune processes. It presents with neuropsychiatric features such as fever, altered mental status, behavioral changes, cognitive dysfunction, seizures, focal deficits, and sometimes autonomic instability. In some cases, the meninges are also involved, resulting in meningoencephalitis.Infectious CausesInfectious encephalitis is most commonly viral but can also result from bacterial, fungal, or parasitic...
27
Encephalitis ll: Pathophysiology
30
Encephalitis is inflammation of the brain parenchyma caused by direct viral invasion or immune-mediated mechanisms triggered by infections or tumors. Both processes lead to neuronal injury, disrupted neurotransmission, and diverse neurological symptoms, often with overlapping clinical and pathological features.Autoimmune EncephalitisIn autoimmune encephalitis, antibodies target neuronal antigens on cell surfaces, synapses, or within neurons. A key example is anti-NMDAR encephalitis, which can...
30


