Background

Spontaneous episodic vertigo syndromes, namely vestibular migraine (VM) and Menière’s disease (MD), are difficult to differentiate, even for an experienced clinician. In the presence of complex diagnostic information, automated systems can support human decision making. Recent developments in machine learning might facilitate bedside diagnosis of VM and MD.

Methods

Data of this study originate from the prospective patient registry of the German Centre for Vertigo and Balance Disorders, a specialized tertiary treatment center at the University Hospital Munich. The classification task was to differentiate cases of VM, MD from other vestibular disease entities. Deep Neural Networks (DNN) and Boosted Decision Trees (BDT) were used for classification.

Results

A total of 1357 patients were included (mean age 52.9, SD 15.9, 54.7% female), 9.9% with MD and 15.6% with VM. DNN models yielded an accuracy of 98.4 ± 0.5%, a precision of 96.3 ± 3.9%, and a sensitivity of 85.4 ± 3.9% for VM, and an accuracy of 98.0 ± 1.0%, a precision of 90.4 ± 6.2% and a sensitivity of 89.9 ± 4.6% for MD. BDT yielded an accuracy of 84.5 ± 0.5%, precision of 51.8 ± 6.1%, sensitivity of 16.9 ± 1.7% for VM, and an accuracy of 93.3 ± 0.7%, precision 76.0 ± 6.7%, sensitivity 41.7 ± 2.9% for MD.

Conclusion

The correct diagnosis of spontaneous episodic vestibular syndromes is challenging in clinical practice. Modern machine learning methods might be the basis for developing systems that assist practitioners and clinicians in their daily treatment decisions.

中文翻译：

---

开发和验证用于诊断和区分自发性发作性眩晕综合征的分类算法：DizzyReg患者注册中心的结果。

背景

自发性发作性眩晕综合征，即前庭偏头痛（VM）和梅尼埃病（MD），即使对于有经验的临床医生也很难区分。在存在复杂的诊断信息的情况下，自动化系统可以支持人工决策。机器学习的最新发展可能会促进VM和MD的床旁诊断。

方法

这项研究的数据来自德国眩晕与平衡障碍中心的预期患者登记处，该中心是慕尼黑大学医院的专门三级治疗中心。分类任务是将VM，MD与其他前庭疾病实体区分开。使用深度神经网络（DNN）和增强决策树（BDT）进行分类。

结果

包括1357例患者（平均年龄52.9，SD 15.9，女性54.7％），MD为9.9％，VM为15.6％。DNN模型对VM的精度为98.4±0.5％，精度为96.3±3.9％，灵敏度为85.4±3.9％，精度为98.0±1.0％，精度为90.4±6.2％，灵敏度为MD的为89.9±4.6％。BDT对VM的精度为84.5±0.5％，精度为51.8±6.1％，灵敏度为16.9±1.7％，对MD的精度为93.3±0.7％，精度为76.0±6.7％，灵敏度为41.7±2.9％。

结论

自发性发作性前庭综合征的正确诊断在临床实践中具有挑战性。现代机器学习方法可能是开发系统的基础，该系统可以帮助从业者和临床医生进行日常治疗决策。