Assessment of deep learning techniques for bone fracture detection under neutrosophic domain

Doaa El-Shahat; Ahmed Tolba

Authors

Doaa El-Shahat Faculty of Computers and Informatics, Zagazig University, Shaibet a Nakareyah, Zagazig, Ash Sharqia Governorate, 44519, Egypt
Ahmed Tolba Faculty of Computers and Informatics, Zagazig University, Shaibet a Nakareyah, Zagazig, Ash Sharqia Governorate, 44519, Egypt

Keywords:

Deep Learning; Neutrosophic Set; Bone Fracture Detection; Artificial Intelligence.

Abstract

With the increasing strain on the health system, there is a growing need for automatic medical
image diagnosis. Emerging technologies for medical diagnosis can help to achieve the goals of
sustainable development. However, analyzing medical images can be challenging due to
uncertain data, ambiguity, and impreciseness. To address this issue, we have developed a novel
BoneNet-NS technique to classify fractures in X-ray bone images. The proposed approach is
based on the power of deep learning (DL) and neutrosophic set (NS) to deal with aleatoric
uncertainty. Moreover, we present two frameworks for integrating NS with DL models:
BoneNet-NS1 and BoneNet-NS2. We employ various DL models, including Xception,
ResNet52V2, DenseNet121, and customized CNN to evaluate both frameworks. Furthermore,
4924 X-ray bone images are utilized to distinguish between fractured and non-fractured classes.
The statistical analyses demonstrate that BoneNet-NS2 performs better than BoneNet-NS1 for
most DL models. Specifically, using the ResNet52V2 model, our proposed BoneNet-NS2
achieved the highest accuracy, log loss, precision, recall, F1-score, and AUC with values of
99.7%, 0.006, 99.7%, 99.7%, 99.7%, and 99.7%, respectively.

DOI: 10.5281/zenodo.11479699