Scholarlink

Background <p>Distinguishing dangerous from benign vertigo remains a diagnostic challenge. Our study aimed to develop and evaluate a machine learning model to differentiate between dangerous and benign vertigo in the outpatient setting across two medical institutions that used different equipment, testing methods, and protocols.</p> Methods <p>Patients in the dizziness clinics of Shuang Ho Hospital (SHH) and Taichung Tzu Chi Hospital (TTCH) were classified into “benign vertigo” and “dangerous vertigo” groups. The data of video-oculography (VOG) and video head impulse test were combined and preprocessed by applying the min–max scaler and the synthetic minority oversampling technique. The optimized Random Forest (RF)–Adaptive Boosting (AdaBoost) model was employed for classification. The SHapley Additive exPlanations method provided feature importance. Model performance was evaluated using a random 70/30 stratified split across sites, and 95% confidence intervals (CIs) were computed using stratified bootstrapping.</p> Results <p>A total of 294 patients were enrolled, including 184 (171 with benign vertigo and 13 with dangerous vertigo) in SHH and 110 (63 with benign vertigo and 47 with dangerous vertigo) in TTCH. The optimized RF–AdaBoost model was used to identify dangerous vertigo. The accuracy was 97% (95% CI: 92 – 100%), the sensitivity was 92% (95% CI: 75 – 100%), the specificity was 97% (95% CI: 93 – 100%), and the AUC was 0.95 (95% CI: 0.86 – 1.00).</p> Conclusion <p>Our VOG-based machine learning model for differentiating dangerous from benign vertigo demonstrated good accuracy and showed potential for use across different equipment, testing methods, and institutional protocols.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated vertigo diagnosis using video-oculography and artificial intelligence for clinic-based triage: Preliminary results

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Yi-Jing Lai,
  • Ching-Fu Wang,
  • Chun-Chen Yang,
  • Chin-Hsun Huang,
  • Chih-Chung Chen,
  • Ting-Yi Lee,
  • Hsun-Hua Lee,
  • I.-Hsien Chen,
  • Tzu-Pu Chang

摘要

Background

Distinguishing dangerous from benign vertigo remains a diagnostic challenge. Our study aimed to develop and evaluate a machine learning model to differentiate between dangerous and benign vertigo in the outpatient setting across two medical institutions that used different equipment, testing methods, and protocols.

Methods

Patients in the dizziness clinics of Shuang Ho Hospital (SHH) and Taichung Tzu Chi Hospital (TTCH) were classified into “benign vertigo” and “dangerous vertigo” groups. The data of video-oculography (VOG) and video head impulse test were combined and preprocessed by applying the min–max scaler and the synthetic minority oversampling technique. The optimized Random Forest (RF)–Adaptive Boosting (AdaBoost) model was employed for classification. The SHapley Additive exPlanations method provided feature importance. Model performance was evaluated using a random 70/30 stratified split across sites, and 95% confidence intervals (CIs) were computed using stratified bootstrapping.

Results

A total of 294 patients were enrolled, including 184 (171 with benign vertigo and 13 with dangerous vertigo) in SHH and 110 (63 with benign vertigo and 47 with dangerous vertigo) in TTCH. The optimized RF–AdaBoost model was used to identify dangerous vertigo. The accuracy was 97% (95% CI: 92 – 100%), the sensitivity was 92% (95% CI: 75 – 100%), the specificity was 97% (95% CI: 93 – 100%), and the AUC was 0.95 (95% CI: 0.86 – 1.00).

Conclusion

Our VOG-based machine learning model for differentiating dangerous from benign vertigo demonstrated good accuracy and showed potential for use across different equipment, testing methods, and institutional protocols.