A Machine Learning Based Quantitative Data Analysis for Screening Skin Diseases Based on Optical Coherence Tomography Angiography (OCTA)

Ji, Yubo; Yang, Shufan; Zhou, Kanheng; Li, Chunhui; Huang, Zhihong

Abstract

Lack of accurate and standard quantitative evaluations limit the progress of applying the OCTA technique into skin clinical trials. More systematic research is required to investigate the possibility of using quantitative OCTA techniques for screening skin diseases. This prospective study included 88 participants (60 normal and 28 abnormal skin samples). In total, 40 OCTA quantitative parameters (3 for epidermis feature, 27 for dermis feature, 10 for vascular feature) were obtained by each OCT and OCTA data volumes. The proposed method relies on linear support vector machines (SVM), while the coefficient of multiple linear regression is also employed to select seven most significant features. Result shows that the proposed method can improve the classification accuracy which can arrive at 93%. Moreover, selected features provide us with direction to determine which biomarker is potential for clinical diagnosis of specific skin abnormalities.

Citation

Ji, Y., Yang, S., Zhou, K., Li, C., & Huang, Z. (2021). A Machine Learning Based Quantitative Data Analysis for Screening Skin Diseases Based on Optical Coherence Tomography Angiography (OCTA). In 2021 IEEE International Ultrasonics Symposium (IUS). https://doi.org/10.1109/IUS52206.2021.9593642

• Conference Name: 2021 IEEE International Ultrasonics Symposium
• Conference Location: Xi'an, China [Online]
• Start Date: Sep 11, 2021
• End Date: Sep 16, 2021
• Acceptance Date: Jun 16, 2021
• Online Publication Date: Nov 13, 2021
• Publication Date: 2021
• Deposit Date: Jun 17, 2021
• Publisher: Institute of Electrical and Electronics Engineers
• Series ISSN: 1948-5727
• Book Title: 2021 IEEE International Ultrasonics Symposium (IUS)
• DOI: https://doi.org/10.1109/IUS52206.2021.9593642
• Keywords: OCTA, Skin disease, Linear Regression, SVM, Machine Learning
• Public URL: http://researchrepository.napier.ac.uk/Output/2781028