Cerebral artery occlusion screening system for emergency medicine with autoencoder-based inference using pulse waves

Cerebral artery occlusion is a leading cause of stroke and death and often results in severe sequelae. Its prompt diagnosis and treatment are crucial for saving lives. We are developing a cerebral artery occlusion screening system usable in ambulances, which consists of a pulse wave measurement device and an occlusion inference method. Toward the ultimate goal that our system is medically and legally approved, we propose a novel occlusion inference method. The pulse wave data is quite small because our device is the world’s first, and its use is currently restricted before approval. Therefore, the present objective is to demonstrate that the proposed method works well even for such small data. The proposed method consists of an Autoencoder (AE) and a Threshold-based Classifier (TC). AE learns how to reconstruct the pulse waves of healthy subjects, and so it cannot reconstruct those of occlusion patients well. TC infers the existence of occlusion by thresholding this reconstruction performance. The proposed method achieves an accuracy of 62 [%] and an F value of 73 [%], outperforming the previous method by 15 [%] in F value. Our datasets are available from the corresponding author on reasonable request.