Enhancing the thermographic diagnosis of maxillary sinusitis using deep learning approach

ABSTRACT

Recent studies witnessed the potential benefits of using deep learning (DL) based approaches in drug discovery, disease diagnosis, treatment planning and many more applications. In addition to software approaches, researchers and clinicians are substantially focusing to standardise the thermography in the medical realm due to its non-contact and non-ionising nature. The promising results of using thermography for sinusitis detection have been remarked in many studies, which have given new directions to conduct further research. In this study, the DL-based approach has been implemented to enhance the thermography-based diagnosis of various classes of maxillary sinusitis, including 1080 thermograms (Dataset: EIE, SLIET, India). More specifically, the study is comprised of three steps: (a) Pre-processing: face extraction using instance segmentation; (b) Classification: four different classes using Inception Network V3 model; and (c) performance evaluation. The major highlight of this study is the classification into four classes: control, unilateral left maxillary sinusitis (ULMS), unilateral right maxillary sinusitis (URMS), and bilateral maxillary sinusitis (BMS). Consequently, an overall accuracy of 99.5% with 99% sensitivity has been achieved by applying the proposed approach. Furthermore, the integration of automatic segmentation of facial features augments the efficiency and accuracy of the diagnostic process by filtering out the unwanted data.

KEYWORDS:

• Maxillary sinusitis
• thermography
• deep learning
• R50-FPN
• inception network V3

Acknowledgements

The authors acknowledge the cooperation of all the participants in data acquisition.

Disclosure statement

No potential conflict of interest was reported by the author(s).