Classifications of tablets and pills by the shape through unsupervised learning

초록

Medication error is one of the most common types of adverse events hap-pening in healthcare. One of the root causes of medication error is that too many drugs share similar characteristics in their names, shapes, and colors. However, the humans’ perceptual structure of images is not clearly identified yet. This study tried to apply unsupervised learning techniques to discovering the possible underlying structure from drug images. Totally 1,500 images selected from 15,080 images were analyzed and the result suggested three dimensional structures of brightness, color and arrangement of images. Drug images are allocated into four clusters: big tablet, white circular, dark colored oval, and colored mixture. Further analysis for capturing shape characteristics is needed for better classification. © 2020, ICIC International. All rights reserved.

키워드