Background and objective

The paper presents a novel technique for the visualisation and measurement of anthropometric features from patients with severe musculoskeletal conditions. During a routine postural assessment, healthcare professionals use anthropometric measurements to infer internal musculoskeletal configuration and inform the prescription of Custom Contoured Seating systems tailored to individual needs. Current assessment procedures are not only time consuming but also do not readily facilitate the communication of musculoskeletal configuration between healthcare professionals nor the quantitative comparison of changes over time. There are many techniques measuring musculoskeletal configurations such as MRI, CT or X-ray. However, most are very resource intensive and do not readily lend themselves to widespread use in, for example, community based services. Due to the low volume of patient data and hence small datasets modern machine learning techniques are also not feasible and a bespoke solution is required.

Methods

The technique outlined in this paper uses physics simulation to visualise the orientation of the pelvis and femurs when seated in a custom contoured cushion. The input to the algorithm is a body shape measurement and the output is a visualised pelvis and femurs. The algorithm was tested by also outputting a multi-label classification of posture (specific to the pelvis and femurs).

Results

The physics simulation has a classification accuracy of 72.9% when labelling all 9 features of the model; when considering 6 features (excluding rotations about the x-axis) the accuracy is increased to 92.8%.

Conclusions

This study has shown that a mechanical shape sensor can be used to capture the unsupported seated posture of an individual during a clinic. The results have demonstrated the potential of the physics simulation to be used for anthropometric feature extraction from body shape measurements leading to a better posture visualization. Capturing and visualising the seated posture in this way should enable clinicians to more easily compare the effects of clinical interventions over time and document postural changes. Overall, the algorithm performed well, however, in order to fully evaluate its clinical benefit, it needs to be tested in the future using data from patients with severe musculoskeletal conditions and complex body shapes.

中文翻译：

---

使用物理模拟从人体形状测量获得3D姿势可视化，以确定坐姿的骨盆和股骨的方向

背景和目标

本文提出了一种新技术，用于可视化和测量患有严重肌肉骨骼疾病的患者的人体测量特征。在常规的姿势评估中，医疗保健专业人员使用人体测量法来推断内部肌肉骨骼结构，并根据个人需求量身定制定制轮廓座椅系统的处方。当前的评估程序不仅耗时，而且不容易促进医疗保健专业人员之间肌肉骨骼结构的交流，也不能随时间变化的定量比较。有许多测量肌肉骨骼结构的技术，例如MRI，CT或X射线。但是，大多数资源非常耗费资源，因此无法轻易地在例如基于社区的服务中广泛使用。现代机器学习技术也不可行，因此需要定制解决方案。

方法

本文概述的技术使用物理模拟来可视化坐在定制轮廓垫上的骨盆和股骨的方向。该算法的输入是人体形状测量，输出是可视化的骨盆和股骨。还通过输出姿势的多标签分类（特定于骨盆和股骨）对算法进行了测试。

结果

标注模型的所有9个特征时，物理模拟的分类精度为72.9％；当考虑6个特征（不包括绕x轴的旋转）时，精度提高到92.8％。

结论

这项研究表明，在诊所期间，可以使用机械形状传感器捕获个人不受支撑的坐姿。结果表明，物理模拟的潜力可用于从人体形状测量中提取人体特征，从而获得更好的姿势可视化。以这种方式捕获并可视化坐姿，应使临床医生能够更轻松地比较随时间变化的临床干预效果并记录姿势变化。总体而言，该算法表现良好，但是，为了充分评估其临床益处，将来需要使用来自具有严重肌肉骨骼疾病和复杂体形的患者的数据进行测试。