10:30   Lower extremities & Motion I Chair: Dennis Janssen Minimally invasive sacroiliac joint fusion revision using a patient specific surgical guide for implant placement A.D. Smelt, J.M. Nellensteijn, N.F.B. Kampkuiper, M. Haenen, M. Leemans, E.E.G. Hekman, F.F. Schröder, M.A. Koenrades Abstract: Objective: Minimally invasive surgical revision of sacroiliac joint fusion (SIJF) requires accurate placement of implants to reobtain or increase stability after implant loosening, implant malposition or pelvic fracture to relief pain. Current management using fluoroscopy and standard instruments challenges accurate placement and does not allow non-parallel implant configurations, which is desirable due to limited space for additional implants. The objective of this study was to develop and evaluate the use of patient specific surgical guides in SIJ fusion revision. Methods: A patient specific guide (PSG) was developed to be fixated in in situ implants or sacroiliac screws by use of Kirschner wires, not requiring any bone seating, and to accurately guide placement of additional implants (iFuse Implant System, SI-Bone, Santa Clara, CA, USA). PSGs comprised a 3D printed main body (polyamide 12, powder bed fusion) and guidance tubes (polyamide 12, fiberglass or stainless steel). Thirteen patients (all female, median age 52 years) underwent SIJ fusion revision in a prone position lateral approach from 2021 to 2024 using a PSG. One to three additional implants were placed resulting in a total of 22 implants. Positional and angular deviations were evaluated from postoperative CTs. Results: All implants were successfully placed without malpositioning complications. A mean total 3D positional deviation of 6.2 ± 2.0 mm (polyamide 12: 7.9-8.3 mm; fiberglass 5.9 ± 1.9; stainless steel: 6.1 ± 2.3) and a total mean 3D angular deviation of 4.2 ± 2.3⁰ (polyamide 12: 1.8-2.0⁰; fiberglass: 5.1 ± 2.7⁰; stainless steel: 3.7 ± 1.6⁰ ) was found. Conclusion: This study suggests that PSGs can facilitate accurate implant placement in minimally invasive SIJ fusion revision even with trans-articular non-parallel implant configurations. This may improve clinical outcome after revision with adequate pain relief and reduce malpositioning and implant loosening complications. Biomimetic Calcium Phosphate Bone Substitutes: Preclinical Development and First-in-Human Clinical Trial Results Yuelian Liu, Yuanyuan ACTA Sun Abstract: Biomimetic Calcium Phosphate Bone Substitutes: Preclinical Development and First-in-Human Clinical Trial Results A protocol and software to assess coil-head stability during TMS gait studies Raven Huijberts, Jennifer Davies, Sjoerd Bruijn Abstract: Transcranial magnetic stimulation (TMS) is a non-invasive method to investigate connections between the cortex and muscles. Despite its potential, TMS is primarily used in static positions due to the precision required for cortical targeting. Current studies using TMS during walking primarily rely on visual assessment or assume coil stability. In these studies, the coil is often kept in place by a harnesses, helmet, or cap. To advance research on gait control, coil stability needs to be validated and standardised. This study aims to standardise coil stability assessment and demonstrate how it can be maintained during normal and perturbed gait. Fifteen participants completed ~20 minutes of normal and perturbed treadmill gait while receiving random TMS pulses. The TMS coil was helmet-mounted and supported by a spring frame to allow natural movement. Before the trials, a static recording of the coil centre’s position and four helmet markers was made. During the trials, three additional face markers were placed. A custom made openly source software package was developed which utilizing these recordings to calculate coil stability. Our results show an average displacement of 0.54 - 1.09 mm across movement directions during normal gait, and 0.71 -1.59 mm during perturbed gait. Peak-to-peak displacements were maximum 10.7 mm during normal gait, 15.9 mm during perturbed gait. However, most stimulations happened within a 3 mm displacement threshold (normal: 91.5% ± 7.7%, perturbed: 80.8% ± 18.9%). Our software enables standardised coil stability assessment. Using this, we demonstrate that with a suitable setup, coil stability can be maintained during normal and even perturbed gait, paving the way for future research on cortex-muscle connections during gait. Estimation of Lower Body Joint Angles with a Minimal IMU Sensor Setup via Transfer Learning on Different Activities Xiaowen Song, Frank Wouda, Jasper Reenalda, Peter Veltink Abstract: Lower body kinematics plays an important role in preventing sports-related injuries by providing information on joint movements and identifying abnormalities. However, traditional motion capture systems require numerous markers or sensors for comprehensive full body motion analysis, making them less feasible for real-world applications. To address these limitations, reduced sensor setups have been considered a more efficient, cost-effective alternative. Moreover, existing deep learning models used to aid sensor reduction experience a decline in accuracy when applied to new, unseen activities, requiring substantial data collection for each new activity. However, gathering large datasets for each new activity is both time-consuming and costly. In this research, we first train a model on available running data at different speeds, with input from only three IMU sensors located at the pelvis and both lower legs, to enhance its generalization ability for new activities. By applying a hybrid Convolutional Neural Network and Long Short-term Memory model, we aim to capture both spatial and temporal patterns in lower body motion. The output of the model includes joint angles for the hips, knees and ankles. For transfer learning, we fine-tuned the pre-trained model on just 6% of all the data specific to 16 different activities, which include gait, activities of daily living and sports. When compared to directly applying the pre-trained model to new activities without fine-tuning, we observe an average reduction in joint angles RMSE from 34.5° to 13.4°. Comparing to training from scratch on these new activities, we find an average decrease in RMSE of joint angles around 2°. These findings highlight the effectiveness of transfer learning in adapting running-based models to other activities with minimal data, emphasizing its potential for broader applications in motion analysis. The minimal sensor setup, combined with transfer learning, presents a promising solution for cost-effective lower body kinematics analysis, offering valuable implications for easier ambulatory measurement. Dichroism sensitive photoacoustic imaging for the assessment in depth of fibers orientation in biological tissue Camilo Cano, Marc van Sambeek, Richard Lopata, Min Wu Abstract: Photoacoustic imaging is an emerging image modality that provides optical contrast while leveraging low acoustics attenuation to reach deeper than conventional optical techniques. Similarly to how optical absorption is used to differentiate materials with spectroscopy analysis, polarization analysis can provide information on the sample anisotropy in magnitude and orientation, which can provide essential information for the study of tissue remodeling. Previous studies have shown the viability of dichroism-sensitive photoacoustic imaging (DS-PAI) for the superficial assessment of tissue anisotropy. In this work, we further develop the model for the imaging in depth of the orientation of fibrous tissue considering the effect of fluence. The method is validated in a tendon sample and contrasted with Monte Carlo simulation. We employ a modified B-Scan DS-PAI system that allows the rotation of the light plane of polarization using a motorized waveplate retarder. According to Malus’ law, the DS-PA signal will display a sinusoidal modulation as a function of the angle between the plane of polarization and the principal axis of the tissue. In this process, fluence modulation also occurs due to the absorption anisotropy in the tissue. Our model shows that fluence modifies the phase of the photoacoustic signal depending on the ratio between the normalized modulations of the fluence and the absorption. By determining the phase of the DS-PA signal through Fourier analysis and correcting it with our model, we can determine the local orientation of the fiber structure in depth. The model was validated on DS-PA (Monte Carlo) simulations and measurements of porcine tendons (1.6 mm thickness) illuminated perpendicularly to its fibers. Results demonstrate that DS-PAI can provide in-depth information on the orientation of fibrous structures in biological tissue. The proposed model describes the behavior of the DS-PA signals and illustrates the effect of fluence and its relevance for a meaningful volumetric assessment of DS-PA signals. In experiments with stacked tendons, we reached an imaging depth of 3.2 mm with a sufficient SNR for processing, demonstrating the technique's potential to image beyond the limit of purely optical methods. Body center of mass Acceleration Estimation From a Wrist-Worn Inertial Measurement Unit During Natural Walking Using Physiological-model-based Neural Networks Shuhao Que, Frank Wouda, Peter Veltink, Ying Wang Abstract: The body center of mass (CoM) acceleration is a highly-valuable parameter for energy expenditure estimation. Smartwatches nowadays have become a common accessory for people to wear on a daily basis, which usually incorporate an inertial measurement unit (IMU) that measures both linear acceleration and angular velocity. It is meaningful to investigate the technical feasibility of estimating the CoM acceleration using wrist-worn IMU measurements to accurately track energy expenditure in daily life for healthy weight management. In this initial work, we first estimated pelvis acceleration, which is closely related to the CoM acceleration, from wrist during fixed-speed natural walking (characterised by cyclic gait) in our collected dataset of 10 human subjects. 2 IMUs were attached to the left wrist and pelvis of each subject. The measured pelvis acceleration was considered the surrogate for CoM. We proposed physiological-model-based long-short term memory (PMB-LSTM) neural network and compared its estimation performance with polynomial kernel regression (PKR). Both methods performed a data-driven mapping from wrist IMU data to pelvis acceleration. The PMB-LSTM is composed of two components, a simplified kinematic model characterized by 27 parameters and one LSTM layer. They were trained in a sequential manner. Leave-one-subject-out cross validation was applied for the performance evaluation with two metrices: root-mean-squared error (RMSE) and its error range (ER). The PMB-LSTM achieved a mean RMSE (ER) value of 0.577 m/s2 (7.1 %) and 0.898 m/s2 (12.5 %), while the PRK achieved a mean RMSE (ER) value of 0.700 m/s2 (8.5 %) on the training data and 0.901 m/s2 (13.6 %) on the test data, respectively. In addition, PMB-LSTM’s kinematic model component with the trained 27 parameters achieved a mean RMSE (ER) value of 1.100 m/s2 (13.9%) on the training data and 1.009 m/s2 (15.4 %) on the test data. Future work will extend to direct CoM acceleration estimation and to other daily life physical activities characterised by non-cyclic gait.

end %-->