کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Clinical Risk Score: A Novel Evaluation Metric for Stroke Classification Systems

نویسندگان :

Morteza Ahmadinasab¹ Amir Reza Attari² Mohammad Saeed Majedi³ Meqdad Taheri⁴

1- دانشگاه فردوسی مشهد 2- دانشگاه فردوسی مشهد 3- دانشگاه فردوسی مشهد 4- دانشگاه فردوسی مشهد

کلمات کلیدی :

Stroke Classification،Microwave Sensing،Clinical Risk Score،Evaluation Metrics

چکیده :

Rapid and accurate classification between ischemic and hemorrhagic stroke is critical because their treatments are completely different. The administration of tissue plasminogen activator (tPA) — the standard treatment for ischemic stroke —is strictly contraindicated for hemorrhagic patients, as it exacerbates bleeding and can be fatal. Developing automated diagnostic systems faces two main challenges: statistically, the prevalence of ischemic cases biases models toward the majority class, and clinically, error costs are non-uniform— mistaking a hemorrhage for an ischemia is life-threatening, while the reverse error is less severe. This study critiques conventional metrics for failing to capture these safety priorities and introduces a new metric called the Clinical Risk Score (CRS) to address this issue. Using S-parameter data from realistic head phantom simulations, we benchmark a Support Vector Machine (SVM) optimized for accuracy against a cost-sensitive SVM optimized for the CRS metric. The results show that while the model with high conventional accuracy achieves good statistical scores, it misses a significant number of hemorrhagic cases. In contrast, the CRS-based approach shifts the decision boundary to protect the high-risk class, reducing life-threatening diagnostic errors by approximately 17%. We conclude that in medical diagnostic systems, optimization must prioritize patient safety over conventional accuracy metrics.