Oral Presentations

TL;DR

This paper presents a convolutional neural network for detecting hip dysplasia in radiographs and discusses VR and 3D-printed tools for surgical training.

Contribution

A CNN model for hip dysplasia detection and a 3D-printed surgical simulator with cortico-medullary differentiation for hand fracture training.

Findings

The CNN model showed potential for accurate hip dysplasia detection using LCEA and Ai measurements.

3D-printed simulators improved tactile feedback for surgical training and were upcycled to reduce waste.

VR simulation reduced anxiety and increased confidence in surgical trainees compared to traditional methods.

Abstract

Early detection of hip dysplasia is pivotal for optimal patient management, yet diagnosis is often delayed due to the subtlety of radiographic signs. This research introduces a convolutional neural network (CNN) model, harnessing the power of artificial intelligence (AI), with the aim to refine and enhance diagnostic accuracy in detecting adult hip dysplasia from plain film radiographs.

This study utilises 1,000 anterior-posterior pelvic radiographs, partitioned into training (700), validation (100), and testing (200) datasets. Through the Unity software, each radiograph underwent systematic manual labelling of essential bony landmarks by multiple assessors, laying the groundwork for the CNN training phase. Our CNN focuses on reliably computing the lateral centre edge angle (LCEA) and acetabular index (Ai) measurements, and subsequently categorising radiographs into normal or pathological, based on established thresholds (LCEA <25 and Ai >10).

Preliminary phases, from data acquisition to training the CNN, have been executed. The potential benefits of this approach include consistent, rapid evaluations with reduced diagnostic variability. Key performance metrics, such as accuracy, precision, recall, F1 score and AUC-ROC, will gauge the model's clinical applicability.

This proof of concept underscores the potential of CNNs in revolutionizing the early detection of adult hip dysplasia. Early-phase testing has yielded positive results, indicating that this model will serve as a valuable adjunct or preliminary screening tool in clinical settings. As such, we believe our CNN model sets the stage for the next chapter in orthopaedic imaging diagnostics.

Standard metacarpal training simulators consist of homogenous plastic and lack the cortico-medullary differentiation which provides the tactical feedback necessary for (a) drilling and (b) to enable K-wire insertion. We are a team of doctors and engineers who set out to develop a new training tool to address these limitations and improve the quality of hand fracture fixation training nationally, and to avoid sending our used single-use simulators to landfill.

We used a stereolithography 3D printer (Formlabs Form 3) to generate the hollow metacarpal cortex. This technique involves depositing liquid resin into the 3D printer with a light source curing the resin into hardened plastic. Medullas were then injection filled.

Prototypes were tested by registrar and consultant colleagues across two UK centres (North Bristol and Royal Free London NHS Foundation Trusts). Feedback on fidelity of metacarpals was collected via online and paper forms.

Products were pitched to an upcycling firm, who take used plastics and turn them into benches and garden furniture. We have collaborated with this firm to take our used simulators and upcycle them into building material.

All hand surgeons agreed that our cortico-medullary differentiated prototypes would enhance surgical training. On average our prototypes scored 78% for replicating tactile feedback when drilling, with 100% agreement that it was at least as good as control devices. Testers agreed that our prototypes represented promise for intramedullary device and K wire training.

We have developed promising prototypes for enhanced hand fracture training to share with courses while avoiding landfill.

The use of virtual reality (VR) technology in healthcare has been increasing over recent years and it has been shown to decrease pain and anxiety during a wide variety of medical procedures.

Over a 1-month period, the Welsh Centre for Burns and Plastic Surgery (WCBPS) trialled the use of the DR.VR Junior, an immersive VR system comprising of a Pico VR headset and optional noise-cancelling headphones (designed and provided by Rescape Innovation). It offers a number of ‘experiences’ in which the wearer is able to have a 360° view of different environments with an accompanying audio description. The aim was for paediatric patients to use the headset as distraction during dressing changes and while they waited for review, with the intention of reducing pain and anxiety.

There were 20 patients who opted to take part, age range 4–16. The headset was more successful in older, male patients (>8). Younger children had a shorter tolerance, possibly compounded by it being one size and therefore a little too big for the youngest of our cohort. The headset itself was portable, easy to use and clean and on the whole, very well received. Most children said they enjoyed the experience and would like to use it again, although a common negative theme within the feedback was a lack of games to play.

As a department, we intend to re-trial a new VR model currently under development, which includes games, in the near future.

Restricted working hours, increased cancellation of operative-lists and the recent pandemic, have provided challenges to surgical training. Virtual Reality (VR) simulation is emerging as a safe and accessible surgical teaching strategy in these pressured environments. The aim of this study was to compare VR-simulation to traditional learning methods in the acquisition of surgical skills required for a trochanteric femoral nail (TFNa, Synthes) insertion.

In this randomised control study, 28 surgical trainees, recruited from a large tertiary hospital were assigned to 2 groups; VR-training intervention group and traditional-training control group. Both were trained then observed inserting the TFNA system into saw-bone femurs. Participants completed validated Likert-style questionnaires pre training, post training and post TFNA insertion to assess orthopaedic skill acquisition, confidence, stress, anxiety, and acceptability.

The VR intervention group compared to traditional control group reported lower anxiety levels (33% vs 50%) and higher confidence levels (61% vs 84%) post TFN insertion. The VR group felt the simulator steps were more accurate (1.83 (+/− 1.11) vs 0.92 (+/− 1.43)) and felt more prepared to attempt the procedure after simulation (0.66 (+/− 1.6) vs −0.35 (+/− 1.82)). There was no statistical difference in time taken (2547.5ss vs 2395ss, p=0.668) or nail guide-wire attempts (2 for both groups, p=0.355).

This study shows that VR-simulation could provide a more realistic and acceptable learning environment for surgical trainees than reading the traditional operation guides. Surgeons and educators should consider investing in VR simulation to provide a safe learning environment for orthopaedic trainees.

Glenoid fractures account for 10%–48% of scapular fractures. Fractures representing greater than 25% of the articular surface are associated with shoulder instability. Surgical fixation is indicated in these circumstances. This can be challenging given the glenoid size, its relation to the humeral head and proximity of neurovascular structures.

By necessity, fixation instrumentation is of small calibre. The surrounding soft tissues can impede achieving the optimum trajectory of fixation devices. This is compounded by the orientation of the glenoid and position of fracture fragments. We describe a technique for facilitating the positioning of screws during glenoid fixation.

Glenoid fracture fixation can be achieved with narrow-diameter cannulated-screw systems. This employs preliminary non-rigid guide K-wires of between 0.9mm–1.2mm in size, to ensure correct screw positioning. To aid in achieving the correct wire orientation, the wire can be passed through an orange (14 gauge) cannula needle, which has an internal diameter of 1.6mm. The cannula needle can be used to both place and orient the K-wire in the most appropriate position to achieve optimum fixation.

The needlepoint sinks into the bone and, therefore, is less likely to displace during passing of the K-wire. Furthermore, the cannula needle is much more rigid than the K-wire, which allows the surgeon to direct the wire with greater ease. The cannula also acts as a tissue guide, preventing soft tissue entanglement in the rotating wire. This technique we describe is particularly useful given the limited accessibility of the glenoid.