An exploration of diabetic foot osteomyelitis x-ray data for deep learning applications

Document Type

Conference Proceeding

Publication Title

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Volume

14845 LNAI

First Page

30

Last Page

39

Publisher

Springer

School

School of Science

RAS ID

71871

Comments

Abela, B., Masek, M., Abu-Khalaf, J., Suter, D., & Gupta, A. (2024, July). An exploration of diabetic foot osteomyelitis x-ray data for deep learning applications. In International Conference on Artificial Intelligence in Medicine (pp. 30-39). Cham: Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-66535-6_4

Abstract

The features of Diabetic Foot Osteomyelitis (DFO) in medical X-rays are subtle. If more salient features are correlated with the DFO class, the model may become overdependent on these salient features and underperform when they are absent. In this paper, we identify DFO correlated features present in a DFO X-ray dataset that become a source of class differentiation for DL models. The dataset contains 882 dorsoplantar foot X-rays with DFO in the forefoot. To isolate distinct information in the original X-rays, six sub-image sets were created from the original X-ray images. A CNN model was then trained on each sub-image set. In order to identify the impact of correlated features, saliency maps were used in conjunction with the corresponding reports to identify which features of the image were activating the model during classification. The highest performing model achieved an accuracy of 76.8% which suggests the model is proficient at classifying DFO, however, the saliency map outputs indicate that classification is being assisted by features correlated with DFO, such as amputations and ulcers. The results of this investigation suggest that when developing X-ray datasets for DFO or other similar diseases, the effect of co-morbid abnormalities needs to be taken into consideration.

DOI

10.1007/978-3-031-66535-6_4

Access Rights

subscription content

Link to Full Text

Share

 
COinS