Explainable Fuzzy Texture Words Model for Automatic Bone Fracture Identification

Abstract

The examination and interpretation of X-ray images is a time-consuming process that requires manual examination of bones for a fracture diagnosis. In this study, we propose a novel feature extraction and classification model for the automatic detection of bone fractures, named Explainable Fuzzy Texture Words (EFTW). The advantage of EFTW, in comparison with other conventional classification methods is that it is explainable, which means that it can explain why an X-ray image is classified as fracture/non-fracture, depending on the existence of a bone fracture. The proposed model utilizes a novel way to express the textural feature vector extracted from X-rays images, using dictionaries of textural words, based on the similarity between the feature vector and the words of the dictionaries. In addition, fuzzy logic is exploited for the quantification of the similarity, using linguistic values. The proposed model is investigated on a real datasets that include 300 X-ray bone images of upper and lower extremity. The results show that EFTW outperforms several state-of-the-art methods, while providing the advantage of explainability that contributes efficiently to the reduction of diagnostic errors, as well as, the increase of the radiologists' productivity. © 2021 IEEE.