Muscle Type Classification On Ultrasound Imaging Using Deep Convolutional Neural Networks

Ultrasonography is a portable, non-invasive and well-established medical imaging method. Recent findings have shown that Ultrasound images of the Musuloskeletal system can be used to reveal information regarding the age or several pathological conditions. In this work we focus on the problem of classification between four different types of muscles (in the transverse and longitudonal plane) from Musculoskeletal Ultrasound Images recorded from 73 healthy subjects. In order to efficiently identify the complex texture patterns formed from the structure of the muscles, Deep Convolutional Neural Networks (DCNNs) were used in this study. A number of state-of-the-art Convolutional Neural Networks were evaluated and the superior performance of the transfer learning scheme was demonstrated over the classical methods for texture-based image classification. The experimental results indicate that the proposed method can successfully recognize the type of the muscle with 89.44% accuracy while also exhibiting good generalization performance.