Automatic classification of male and female skeletal muscles using ultrasound imaging. (March 2020)

• Record Type: Journal Article
• Authors: Xu, Jingxu; Xu, Depeng; Wei, Qianyue; Zhou, Yongjin
• Is Part Of: Biomedical signal processing and control. Volume 57(2020)
• Publication Date: 2020-03
• Subjects: Ultrasound -- Image -- Skeletal muscle -- Gender difference -- Deep learning; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101731
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Skeletal muscles of males and females have certain differences in metabolism, composition, strength, endurance, and so on. However, it remains unknown if the differences between the two genders can be directly observed or quantitatively calculated from ultrasound images. The ability to extract such information from images has the potential to assist practitioners in precision medicine or the design of improved health plans. This paper proposes a deep learning-based approach to classify the skeletal muscles from different gender groups using ultrasound imaging. Specifically, 1498 ultrasound images, collected from skeletal muscles from 107 young male and female subjects, are used in training and classification tests. The overall classification accuracy achieved is 95.2 %. To understand the abstract features that the neural networks use in the classification task, saliency and occlusion maps are employed. These results reveal there is a large disparity in the distribution of the saliency features for the data from the two genders; the distribution in the images from male subjects is substantially more concentrated. With a special occlusion design customized for this concentrated pattern, the classification accuracy rate for male group is increased to 100.0 %. In conclusion, these preliminary results demonstrate that based on ultrasound images of skeletal muscles, the subjects' gender can be automatically classified.