Design and Validation of a Polyaxial Knee Mechanical Prosthesis Alignment Protocol, Based on a Multivariate Biomechanical Model

• Autores: Andrés Mauricio Cárdenas Torres
• Directores de la Tesis: Alher Mauricio Hernández Valdivieso (dir. tes.), Juliana Uribe Pérez (dir. tes.)
• Lectura: En la Universidad de Antioquia ( Colombia ) en 2022
• Idioma: inglés
• Títulos paralelos:
  • Diseño y Validación de una Protocolo de Alineación de Prótesis Mecánicas Poliaxiales basado en un modelo biomecánico multivariado
• Enlaces
  • Tesis en acceso abierto en: hdl.handle.net
• Resumen

  • The above-knee lower limb loss is a common and complex amputation type, which compromises the loss of two fundamental joints for bipedal walking, thus limiting the person's autonomy. Its etiology is associated with health problems, mainly vascular or traumatic.

    The transfemoral prosthesis is a supportive orthopedic device for the rehabilitation and adaptation of the amputee to their new mobility condition. Typically, the prosthesis is assembled with a socket, suspension system, prosthetic knee or joint unit, pylon, and prosthetic foot.

    The adaptation to the prosthetic device is imperative for the amputee's functionality to improve; therefore, the proper performance of the prosthesis must be ensured. Prosthetic alignment is a procedure for adjusting the prosthesis according to the anatomical and biomechanical conditions of the amputee. The alignment procedure typically is conducted in three stages: bench, static, and dynamic alignment. The bench alignment includes the prosthesis assembling and the components aligning according to the load lines of the prosthesis to lock the joint unit. The purpose of static alignment is to match the prosthesis' load lines to the amputee's anatomical lines to ensure balance during standing. Dynamic alignment studies the amputee's gait to identify deviations associated with prosthetic misalignment; so, a fine adjustment of the prosthesis is performed.

    The prosthetic alignment procedure is considerably dependent on the prosthetist's knowledge and skills in gait analysis and prosthetic alignment, as well as the amputee's communication skills and the rehabilitation center's assistive technologies. This decreases the likelihood of achieving nominal prosthetic alignment, making gait deviations the first sign of prosthetic misalignment. If the gait deviation is sustained over time, the amputee changes his or her gait patterns, resulting in diseases of the amputee's musculoskeletal system.

    Multiple measurement systems are used to analyze the gait of humans. The recording of spatio-temporal, kinetic, kinematic, thermal, and muscle activity parameters provides information to detect gait deviations associated with prosthetic misalignment; however, there are no devices dedicated to alignment assessment. Often prosthetic fitting centers do not have access to gait and standing analysis technology; therefore, subjective strategies such as standing observation, visual gait analysis and verbal feedback from the amputee are used to adjust prosthesis alignment. Under this consideration, prosthetic alignment is highly subjective, so the integrity of the amputee's health is at risk. Computational approaches have been proposed to support transtibial prosthetic alignment procedures; however, the scientific literature does not report computational models dedicated to assess the alignment of transfemoral prostheses, so there is still a delay in this subject.

    In this thesis, we consider that prosthetic gait is a multivariate system in which qualitative and quantitative variables of the patient and prosthesis are affected by prosthetic alignment and their relationship can be recognized with the gait and standing analysis. Under this premise, we proposed to develop a new protocol for the alignment of transfemoral mechanical prostheses supported by computational models for the prosthetist's aid during the prosthetic alignment.

    Initially, a literature review was performed to identify the qualitative and quantitative parameters, and the technologies typically used during the gait analysis and prosthetic alignment procedure. A prevalence was found in the use of kinetic, spatiotemporal, muscle activity, body balance, comfort, and stump temperature parameters for the evaluation of prosthetic gait. Likewise, the study allowed us to recognize the magnitude and direction of the alignment variations, the prosthetic elements aligned, the characteristics of the alignment tests, and the population size for recording. The literature review found a lag in the dynamic alignment research, particularly for transfemoral prostheses.

    Based on the literature review findings, we proposed to record twenty-eight (28) parameters of the Ground Reaction Force (GRF), twelve (12) spatiotemporal parameters, six (6) electromyography parameters of the tibialis anterior, gastrocnemius medialis, rectus femoris, and the biceps femoris, and nineteen (19) stump temperature parameters. Additionally, a seventeen (17) question survey was applied to evaluate the prosthetic comfort, and anthropometric and sociodemographic information of the volunteers was asked. Random alignment variations of the socket were ranged between -18.0° to 28.0° in flexion-extension, adduction-abduction, and internal-external movements. The foot alignment variations ranged between -13.0° to 11.0° in dorsi-plantar flexion, eversion-inversion, and internal-external rotation.

    Five (5) alignments were variated on the prosthesis for each amputee, one (1) nominal alignment and four (4) misalignments. The nominal alignment was judged by a senior prosthetist at the Mahavir Kmina Artificial Limb Center. The alignment test was double-blind, as neither amputees nor prosthetists were aware of the prosthetic misalignments. The alignments order was random. Each amputee walked down a hallway for fifteen (15) minutes and all parameters were recorded.

    All parameters were processed to identify inter-subject statistically significant differences between nominal alignment and misalignments, to find the descriptive variables of the static and dynamic prosthetic alignment procedure. The prosthetic misalignment produces statistical differences of the GRF for the prosthetic limb during walking trials; however, no differences were observed in standing. The misalignments did not produce significant differences in the amputees’ balance, comfort, and muscular activity of the sound limb, during gait and standing trials. Generally, the prosthetic gait of the transfemoral amputees was faster, more unstable, and fatiguing than the normal gait of the control group.

    Parameters associated with the stump temperature did not show significant differences between both alignment conditions; however, the intra-subject analysis of the temperature's variation coefficient was different between nominal and misalignment for more than 70.0% of amputees. The inter-subject analysis of the Ground Reaction Force (GRF) showed statistical differences between both alignment conditions.

    Statistical analysis showed no significant differences between nominal alignment and misalignments during static alignment. This limited the scope of the alignment protocol proposed in this thesis to dynamic alignment. From the statistical analysis, it was identified that the following parameters behave as differentiating descriptors between nominal alignment and misalignment during dynamic alignment: the braking force impulse (I_3), propulsion force impulse (I_4), duration of the stance phase (t_1), duration of the braking phase (t_4), duration of the propulsion phase (t_5 ), time to propulsion peak (t_7 ), time to midstance valley (t_9), the impulse of terminal stance and pre-swing (I_6), the loading rate (LR), the braking (BI_V), and propulsion impulse (PI_V).

    The alignment protocol proposed in this thesis includes two computational models. The Support Machine Vector (SMV) with Gaussian Kernel was used to classify GRF parameters of the amputee’s gait between nominal and misalignments. The dataset was divided into 80% for training and validation, and 20% for model testing. The SVM model separated the dataset between nominal alignment and misalignment with 95.5% accuracy. The confusion matrix shows a 5.5% false-negative rate (FNR) for the misalignment class and a 1.8% FNR nominal class.

    A Bayesian Regularized Artificial Neural Networks with 30 hidden layers was trained to estimate the magnitude and direction of prosthetic misalignment in flexion-extension, abduction-adduction, and internal-external rotation of the socket, and dorsiflexion-plantarflexion, inversion- eversion, and medial-lateral rotation of the prosthetic foot. The model could reproduce 94.11% of the information. The histogram shows 0.51° error for the estimated parameters; therefore, using both models, we propose the alignment protocol.

    The computational alignment protocol was validated in the Mahavir Kmina Artificial Limb Center. Two (2) transfemoral amputees were recruited for the trials. One junior and one senior prosthetist accompany the validation tests. The alignment protocol was iterated a maximum of 3 times, to limit interactions with the amputee by COVID-19 biosafety standards. During each iteration, junior and senior prosthetists evaluated the amputees' gait on a scale from zero (0) to ten (10).

    The nominal alignment of the first amputee was not achieved throughout the three iterations, and prosthetists finally rated the prosthetic gait as 8.0. The prosthetists and computational protocol matched in the misalignment prosthesis for all three iterations. The non-convergence of the nominal alignment could be due to the precision of the prosthetists in adjusting the angles suggested by the computational protocol, and the learning curve in the use of the protocol.

    In the validation session of the alignment protocol for the second amputee recruited, the prosthetists were more skilled in the alignment’s adjustments, so the nominal alignment was achieved in the second iteration, ranging the amputee’s gait at 9.6.

    The prosthetists on average scored 8.18 on the prosthetic gait after applying the computational alignment protocol. The natural amputees' gait patterns affected gait quality; therefore, the correction of gait deviations should be done with a posterior treatment. Prosthetists scored 8.52 for this kind of computational aid of the prosthetic alignment, and they stated that the alignment protocol allowed them to do a better job, rating it with an 8.58. The senior prosthetist stated that the protocol made him take 37 minutes longer than usual and the junior prosthetist stated that computational protocol did not take him longer.

    The result of our computational alignment protocol presents an advance in the study of dynamic prosthetic alignment for transfemoral amputees; however, the prosthetic alignment protocol should continue to be studied to clarify the uncertainties caused by the intersubjectivity of the data and to find new strategies to support prosthetists. The rate of convergence of the protocol could be improved by retraining the computational models with a larger dataset; however, the accuracy of angles adjustment is perhaps affecting the convergence of the nominal alignment. Therefore, further research should be focused on the development of more precise alignment tools.