研究建立了一套基於深度學習的程式,對OCT定點量測橈動脈處之表皮組織圖像作圖像處理,並建立了相關的圖像增強程式集合。最終訓練出2個U-net模型,可以從OCT圖像中分別提取任意時長的連續脈搏波或提取出心搏速率圖。研究者進行了初步的人體實驗,分別在站立,靜坐與平臥三種姿態下定點量測橈動脈處的表皮組織。對於心搏速率圖,研究者計算了mean_nni, sdnn, sdsd, nni_50, pnni_50, nni_20, pnni_20, rmssd, median_nni, range_nni, cvsd, cvnni,共12項時域的心率變異性指標,並進行統計學分析。對於連續的脈搏波,本研究對其進行多尺度熵分析,並對結果作統計分析。
In this study, a set of programs based on deep learning was established for image processing of the epidermal tissue images at the radial artery of the OCT fixed-point measurement, and a set of related image enhancement programs was established. Finally, two U-net models are trained, which can extract continuous pulse waves of any length or heart rate graphs from OCT images. The researchers conducted preliminary human experiments, respectively measuring the epidermal tissue at the radial artery in three postures: standing, sitting, and lying down. For the heart rate chart, the researchers calculated mean_nni, sdnn, sdsd, nni_50, pnni_50, nni_20, pnni_20, rmssd, median_nni, range_nni, cvsd, cvnni, a total of 12 time-domain heart rate variability indicators, and performed statistical analysis. For continuous pulse waves, this study conducts multiscale entropy analysis on it, and makes statistical analysis on the results.