Assessing duchenne muscular dystrophy with force-controlled ultrasound

In this paper, we present a technique for quantitative discrimination of Duchenne Muscular Dystrophy (DMD). Our ultrasound image data is generated with a novel force-controlled ultrasound acquisition system that allows precise ultrasound image acquisition at a predetermined force. We use the texture of ultrasound images, as calculated by the Canny edge detector, as the input image feature for our analysis algorithm. After statistically sieving through the edge detection parameters on our training set, we identify the set of parameters significant within a threshold. Decision trees are then trained on these significant parameters over a training dataset with cross-validation, and evaluated on accuracy, precision, specificity and sensitivity on a separate test dataset. We discuss the performance of our system, by muscle groups, on data collected with our device in a recent clinical study. Using depth of the image as a proxy for image regions, we evaluate the extent to which the performance of our system is robust to region-of-interest selection. Our method holds significant promise for automated assessment of Duchenne Muscular Dystrophy using force-controlled ultrasound image acquisition in a reliable and robust manner.

1. Introduction

Ultrasound is a versatile imaging modality. While freehand ability imparts a degree of convenience, sonographer and patient motion-induced image variability serves as a hindrance to developing standardized diagnostic applications. One of the applications that we foresee is the use of ultrasound to assess and track Duchenne Muscular Dystrophy (DMD) [1]–[4]. DMD is a life-threatening disease that affects children of the male gender, resulting in fatty infiltrations in muscle and eventual death. DMD is evaluated today by using tests like the six-minute walk test [5], which can easily be confounded by other factors that relate to the day of the test, such as the mood of the child. Therefore, there is an urgent need to develop a non-invasive, highly reliable, and repeatable method by which DMD could be tracked. This would be of enormous interest to DMD-related drug trials where measurement of drug impact needs to be characterized in a less-confounding way.