Diabetes Wound Perimeter Analysis Using Pixel Per Metric

Abstract

The study was conducted to test the accuracy of pixel permetric (PPM)-based diabetic wound perimeter using canny edge, gaussian filter, streamlit, OpenCV, Python and aruco marker. K-Means is used to detect, classify and segment three types of wounds namely granulation, necrotics and slough. Images of wounds are taken directly on the patient so the dataset is primary data. Accuracy tests are carried out by comparing manual measurements against digital calculations. Manual measurements use mica as a sketch. The thread is used to shade the mica sketch and the length of the thread will be measured using a ruler. Aruco markers are used as a reference to the length of the object. The results of the study from 7 sample data received an average accuracy error of at least 0.49% and a maximum of 5.75%. This is influenced by various factors including validation of manual measuring results that are still less thorough, sharpness of the image, and calibration of the camera. Of the three types of tissue wounds, granulation is the most identifiable type, followed by slough, and the most difficult to identify is necrosis. The study concludes that the results of the accuracy-test have obtained a value that is in accordance with the problem limit, namely accuracy above 90%, with the independent T-test value homogeneous test is t_hitung< t_tabel equivalent to 0.005535 < 2.228 with a deviation of ꭤ = 0.05 so that it is concluded that there is no significant difference in the two-variable values of manual measurement to digital planimetry measurements. Further research can then test accuracy with artificial intelligence deep learning methods with sample datasets such as uNet, SegNet, and other methods