Automatic Vascular Localization with a Flexible Tactile Sensing Dense-Array

This paper reports a flexible tactile sensing dense-array to locate blood vessels when arterial pulse monitoring or autonomous blood sampling without visual input in an intelligent medical system. With a deep learning neural network, a dense-array with 18 pressure sensing chips for tactile-force detection is able to distinguish 6 diverse body parts on the forearm surface: artery, 2 veins, bone, tendon and muscle. The time-variant tactile mappings are taken as images to train a six-layer convolutional neural network (CNN) and the overall recognition precision of this multi-classification model reaches 93.90%. Thereafter the flexible array is compliantly attached onto a silicone cylinder. While the cylinder is rolling in the target region on the forearm, the real-time vascular localization is thereupon automatically performed, which facilitates the following pulse signal acquisition at the radial artery or blood drawing at the elbow. Compared with vision-based approaches, our tactile strategy requires smaller data size, leading to less computing resources and faster identifications, which are of great concern in wearable applications.