Body Part Diagram Recognition in Medical Records: Application of the Histograms of Oriented Gradients and the Mahalanobis-Distance-Based Classifier
Miho Ohsaki, Hayato Sasaki, Hiroharu Kawanaka, Shigeru Katagiri · 2018
Generally, medical records such as surgery consent forms include body part diagrams on which doctors illustrate the positions, sizes, and conditions of disease. Which kinds of body part diagrams are included is useful information for managing medical records and discovering knowledge from them. This study aims to develop an automatic recognition system of body part diagrams. The conventional study applied a method consisting of the modified weighted direction index histogram and the modified quadratic discriminant function that had a 98.52% recognition rate. We proposed a method consisting of the histograms of oriented gradients and the Mahalanobis-distance-based classifier, with the expectation of better performance and more general and easier processing and parameter setting. In the experiments of which task was to recognize one body part diagram in a medical record, our method outperformed the conventional method. It achieved 100.00% recognition rates for almost all the experimental conditions.