Quantification of Muscle Thickness in Gastrocnemius ultrasound images with deep learning

The automatic extraction of muscle thickness is an important application in the clinical routine that has been studied extensively in recent years. Following this trend, the well-established UNet and other state-of-the-art convolutional neural networks are assessed for automating this task. A two-step approach is presented to extract the muscle thickness. As the first step, different segmentors are incorporated to delineate the deep and superficial aponeurosis effortlessly. Afterwards, the muscle thickness is calculated by taking the average distance between the two aponeuroses at different muscle points. The examined dataset to evaluate this method consists of ultrasound images of gastrocnemius medialis of 116 young and healthy subjects of different sex. Regarding the results, the deep learning architectures used in this study have achieved similar to human- level performance. In particular, an overall discrepancy between the automatic and the manual muscle thickness measurements equal to 0.11 mm is reported, a significant result that demonstrates the feasibility of automating this task. Furthermore, the Bland-Altman analysis of the measurements exhibits no systematic errors since most differences fall between the 95% limits of agreement. Finally, the two readings have a 0.99 Pearson's correlation coefficient (p < 0.001, validation set) and the ICC (2, 1) has surpassed 0.99, showing the reliability of this approach.