Please use this identifier to cite or link to this item:
|Title:||Evaluation of Deep Learning Models for Lower Extremity Muscle Segmentation in Thermal Imaging||Authors:||Ergene, M.C.
|Issue Date:||2023||Publisher:||Springer Science and Business Media Deutschland GmbH||Abstract:||Competition and market size in sports are constantly increasing. In this case, one of the biggest problems of sports clubs is athlete injuries. Especially in football, athlete injury costs are very high. However, most injuries are non-contact and preventable. Sports medicine specialists utilise many medical imaging methods for the prevention of sports injuries. Thermography is an imaging method that has been used in the examination of sports injuries in recent years. Fast and accurate segmentation of muscle regions in thermal images enables more objective analyses. In this study, lower extremity thermal images were taken from football players of a super league club for a certain period of time. From these raw thermal images, 9 different muscle groups of the athletes were labelled and a dataset was created. U-Net, FPN, Linknet and PSPNet segmentation models were trained with this dataset. IoU, F1, Precision, Recall, Precision, Recall evaluation metrics were used to evaluate these models. In the separate models trained for each muscle group, the IoU value achieved over 95% success. When the results of the study are analysed, it is discussed that these segmentation models can be used as a critical tool in injury analysis and evaluation in athletes. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.||Description:||2nd Workshop on Artificial Intelligence over Infrared Images for Medical Applications, AIIIMA 2023 -- 2 October 2023 through 2 October 2023 -- -- 302149||URI:||https://doi.org/10.1007/978-3-031-44511-8_9
|Appears in Collections:||Scopus İndeksli Yayınlar Koleksiyonu / Scopus Indexed Publications Collections|
Show full item record
checked on Dec 4, 2023
Items in GCRIS Repository are protected by copyright, with all rights reserved, unless otherwise indicated.