Reliability of a clinical 3D freehand ultrasound technique: Analyses on healthy and pathological muscles

Highlights

• Simplified approach for extracting relevant clinical outcomes is proposed and evaluated.

• Effectiveness of 3DfUS as a clinical tool for differentiating between healthy and pathological medial gastrocnemius muscles.

• The inherent levels of error provide a useful reference for the wider community using this 3DfUS technique.

• Details of the current software implementation are available in an open-source software library.

Abstract

Background and objective

3D freehand Ultrasonography is a medical imaging technique that can be used to measure muscle and tendon morphological and structural properties, including volume, lengths and echo-intensity. These properties are clinically relevant in neurological disorders such as spastic cerebral palsy to monitor disease progression and evaluate the effect of treatment. This study presents a methodology for extracting these parameters along with a clinical reliability analysis for the data acquisition and processing.

Methods

The medial gastrocnemius muscles and Achilles tendon of 10 typically developing children and 10 children with spastic cerebral palsy were assessed. An open-source in-house software library developed in Python (Py3DFreeHandUS) was used to reconstruct, into one 3D data set, the data simultaneously acquired from an US machine and a motion tracking system. US images were manually segmented and linearly interpolated by means of a new simplified approach which involved sequentially decreasing the total number of images used for muscle border segmentation from 100% to 5%. Acquisition and processing reliability was defined based on repeated measures from different data processers and from different data acquirers, respectively.

Results

When only 10% of the US images were outlined, there was an average underestimation of muscle volume of 1.1% and 1.6% with respect the computation of all the available images, for the typically developing and spastic cerebral palsy groups, respectively. For both groups, the reliability was higher for data processing than for data acquisition. High inter-class correlation coefficient values were found for processing and acquisition reliability, with worst case values of 0.89 and 0.61, respectively. The standard error of measurement, expressed as a percentage of the average volumes, was smaller than 2.6 ml (4.8%) in all cases. Conclusions: The present analysis demonstrates the effectiveness of applying 3D freehand ultrasonography in a clinical setting for analysing healthy and pathological paediatric muscle.

Introduction

Among other contributing factors, muscle morphology is a crucial component influencing muscle function. Parameters such as muscle volume and length can provide insight into abnormal muscle behaviour that may be associated with decreased function [1]. These parameters can be altered in various musculoskeletal and neurological disorders, such as spastic cerebral palsy (SCP) [2], [3]. Previous studies have reported a reduction of muscle volume and length for the medial gastrocnemius (MG) in children with SCP [4], [5]. Muscle composition can also be altered due to an increase in internal fat and fibrous content with a decrease in water particles [6], [7]. All these alterations can be quantified using volumetric medical imaging.

2D ultrasonography (US) is a non-invasive medical imaging modality that can easily be applied to multiple sites of the body [8]. However, 2D US is confined by the width and tissue depth penetration of the US transducer [9], not allowing visualisation of the entire muscle and tendon [10]. 3D freehand US (3DfUS) is a technique capable of coping with this drawback by combing a conventional 2D US system with a pose (position and orientation) sensor attached to the US transducer [11], [12]. This results in volumetric data sets. 3DfUS has recently been gaining popularity due to advancements in software and hardware performance, making it suitable for several clinical applications [13].

3DfUS validation analysis in-vitro and on animals showed promising results [14]. These were also satisfactory when analysed in-vivo for the MG muscle of healthy adults, by showing good agreement with respect the data extracted from magnetic resonance imaging [15], [16]. Furthermore, satisfactory reliability results were found [16]. These parameters were obtained with the lower leg submerged in a water tank, which aids in limiting acquisition errors such as US transducer pressure over the skin. However, the use of a water tank is not convenient for routine use, hence the reported reliability values may not be entirely representative. Routine investigations on lower limb muscles are instead carried out using copious amounts of acoustic streaming gel in between the US transducer and the skin [17]. The copious amount of gel could help in minimising the transducer pressure during the entire acquisition. This is not always possible, as the size and curvature of the leg can influence the amount of acoustic streaming and visibility of the entire muscle border. To date, no investigation has reported the errors introduced when applying 3DfUS on the MG in a routine setting. Moreover, the reliability is yet to be evaluated in pathological populations, where the muscle borders may appear less defined, such as in SCP [16]. This crucial step is needed to define the potential errors when using 3DfUS for clinical purposes.

After computing a 3D reconstruction, muscles are manually outlined using the reconstructed 2D images, and the corresponding borders are interpolated to generate a 3D mesh of the muscle and tendon. A high number of manually outlined images ensures high accuracy, but results in a time-consuming procedure. To reduce this analysis time and make the method more attractive for clinical use, the validity of using only a selection of the acquired images for manual segmentation should be further explored.

The aim of this investigation is to perform a comprehensive reliability analysis of the 3DfUS technique for acquiring and processing muscle volume, length and echo-intensity (EI) of the MG in TD and SCP children, in a clinically applicable manner. In addition, a gradually simplified procedure will be undertaken to determine the optimized linear density of manual segmentations required to accurately compute muscle volume and EI. It is hypothesised that the data acquisition stage introduces more errors than that of data processing, that the measurement errors for morphological muscle properties are lower than the expected differences between pathological and healthy muscles and that a limited number of muscle border segmentations is sufficient to estimate volume and EI.