Insights into the spectrum of transtibial prosthetic socket design from expert clinicians and their digital records

Abstract

Background: Transtibial prosthetic sockets are often grouped into patella tendon bearing (PTB) or total surface bearing (TSB) designs, but many variations in rectifications are used to apply these principles to an individual's personalised socket. Prosthetists currently have little objective evidence to assist them as they make design choices.

Aims: To compare rectifications made by experienced prosthetists across a range of patient demographics and limb shapes to improve understanding of socket design strategies.

Methodology: 163 residual limb surface scans and corresponding CAD/CAM sockets were analysed for 134 randomly selected individuals in a UK prosthetics service. This included 142 PTB and 21 TSB designs. The limb and socket scans were compared to determine the location and size of rectifications. Rectifications were compiled for PTB and TSB designs, and associations between different rectification sizes were assessed using a variety of methods including linear regression, kernel density estimation (KDE) and a Naïve Bayes (NB) classification.

Results: Differences in design features were apparent between PTB and TSB sockets, notably for paratibial carves, gross volume reduction and distal end elongation. However, socket designs varied across a spectrum, with most showing a hybrid of the PTB and TSB principles. Pairwise correlations were observed between the size of some rectifications (e.g., paratibial carves; fibular head build and gross volume reduction). Conversely, the patellar tendon carve depth was not associated significantly with any other rectification, indicating its relative design insensitivity. The Naïve Bayes classifier produced design patterns consistent with expert clinician practice. For example, subtle local rectifications were associated with a large volume reduction (i.e., a TSB-like design), whereas more substantial local rectifications (i.e., a PTB-like design) were associated with a low volume reduction.

Clinical implications: This study demonstrates how we might learn from design records to support education and enhance evidence-based socket design. The method could be used to predict design features for newly presenting patients, based on categorisations of their limb shape and other demographics, implemented alongside expert clinical judgement as smart CAD/CAM design templates.

Keywords: CAD/CAM; PTB; TSB; expert system; knowledge-based system; machine learning; prosthetic limb design.

Figure 1

Data processing from 3D scan of limb and CAD/CAM socket design, extracted rectification design feature locations and sizes, expressed as design variables, and categorised.

Figure 2

Distributions of rectification sizes for sockets described as PTB and TSB designs. “Build” denotes material is added, and “Carve” denotes material is removed.

Figure 3

Example socket designs to PTB and TSB intent, plotted on the residual limb shape. Some training dataset designs had clear PTB or TSB intent, and others lay on a hybrid spectrum between PTB and TSB. Colour key indicates rectification design map in mm. Positive (red) represents carve, and negative (blue) represents build-up.

Figure 4

Three methods of assessing association between the sizes of three example pairs of rectifications. First, a scatter plot (left) of rectification sizes shows common combinations of rectification sizes, where each point represents one of the 163 training sockets. Both variables are continuous. The probability of combinations calculated by Kernel Density Estimation (KDE) is superimposed as a colour map. Circles represent nominally PTB sockets, and crosses are nominally TSB sockets. Three slices through the dataset are then used (centre) to define low, medium and high values of one rectification. For these categories, the corresponding probability density function of the other rectification is plotted. Finally, the Gaussian Naïve Bayes (NB) classifier is used to show the probability that a prosthetist would choose combinations of low, medium and high sizes of each rectification having previously chosen the size of another rectification (right). Results are shown for a highly associated pair (LP and MP, top), an un-associated pair (PT and TC, middle) and a pair which contains different association options (DE and VC, bottom).

Figure 5

By pre-defining the size of volume reduction across the socket, the Naïve Bayes classifier was used to provide probabilities of the clinician choice of size of the other rectifications across the training dataset. With 3 categories, a probability of 33.3% would represent no preference. This shows clear groups of more PTB-biased socket design associated with a decision to perform low volume reduction (top row) and more TSB-biased features associated with a large-sized volume reduction (bottom row). In rectification map, red represents a carve or volume reduction in the socket design, blue represents build, and white is a close match to the limb shape.