Quantitative Texture Analysis and Automated Segmentation of Patellar Tendon Sonographic Images on Collegiate Athletes
Files
TR Number
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Patellar tendinopathy, a common cause of anterior knee pain, is primarily due to degenerative changes in the patellar tendon resulting from repetitive stress. While the tendon itself is the primary focus, damage can also manifest in surrounding tissues, such as the subcutaneous tissue and the fat pad. Traditional diagnostic methods, including clinical evaluation and conventional imaging techniques, have limitations that have led to the development of ultrasound texture analysis as a valuable tool for a more detailed assessment of these tissues. These limitations can also be addressed through machine learning algorithms such as convolutional neural networks to automate image segmentations for determination of quantitative measures to determine injury. The primary objective of this research is to identify texture parameters the distinguish pre-season from post-season states. Additionally, this study aims to partially automate the segmentation process to reduce the time required by trained clinicians, as reviewing premade segmentations is significantly faster than performing manual segmentations. With informed consent and under an IRB-approved protocol, clinical evaluations and ultrasound imaging were performed by Edward Via College of Osteopathic Medicine (VCOM) on 33 Division I collegiate basketball athletes (16 males and 17 females; mean age 19.9 ± 1.4 years). Four imaging sessions were collected over from 2017 to 2019, both pre- and post-season. B-Mode + SWU Images were acquired using a GE LOGIQ S8 (General Electric, USA) ultrasound machine utilizing a 12L probe. Texture analysis of the images was performed using 3D Slicer specifically Radiomics which gives 107 parameters containing first order (16 including mean, median, variance, etc.), second or higher order parameters (Gray level co-occurrence matrix (glcm), Gray level dependence matrix (gldm), Gray level run length matrix (glrlm), Gray level size zone matrix (glszm), Neighbouring Gray tone difference matrix ngtdm) [1], [2]. This research demonstrates that quantifiable differences in texture parameters exist between pre- and post-season states within the subcutaneous tissue, the tendon proper, and Hoffa's fat pad of collegiate athletes. Additionally, the developed automated machine learning segmentation algorithm, after refinement, will assist in injury prevention, facilitate non-invasive injury diagnostics, and serve as a time-saving tool for clinicians.