Skip to main content

Individualized treatment planning in chronic back pain patients by advanced imaging and multi-parametric biomechanical models

Periodic Reporting for period 4 - iBack (Individualized treatment planning in chronic back pain patients by advanced imaging and multi-parametric biomechanical models)

Reporting period: 2019-12-01 to 2020-11-30

Chronic back pain is a major burden and source of disability worldwide. The objective of iBack is to break new ground in individualized therapy planning in these patients based on novel in-vivo imaging and image analysis. We are developing fully automatic approaches for individualized biomechanical models that will allow personalized treatment planning and outcome prediction.
"1. Imaging
a) Computed tomography.
We evaluated the effect of several dose reduction techniques (both on the acquisition and reconstruction side) on BMD, trabecular parameters and results of finite-element (FE) analysis (doi: 10.1371/journal.pone.0159903 doi: 10.1038/srep38441 doi: 10.1007/s00330-017-4904-y doi: 10.1007/s00198-018-4675-6 doi: 10.1007/s00330-019-06090-2). We were able to show that sparse sampling allows for further dose reduction while still delivering robust, quantitative bone mineral density (BMD) values. Most importantly, computed tomography (CT) imaging radiation dose could be reduced substantially to 64% with no impact on strength estimates obtained from FE analysis. To quantify BMD values also in non-dedicated CT exams (""opportunistic screening""), we established different methods to calibrate CT scans retrospectively (doi: 10.1038/s41598-017-17855-4 doi: 10.1097/RCT.0000000000000518) and evaluated different methods for opportunistic screening (doi: 10.1007/s00330-020-07319-1 doi: 10.1007/s00330-019-06263-z doi: 10.1007/s00198-019-04910-1 doi: 10.1007/s00330-019-06018-w).

b) Magnetic Resonance Imaging
In collaboration with the ""Body Magnetic Resonance Research Group"" (PI Dimitrios Karampinos PhD) we developed a Magnetic Resonance Imaging (MRI)-protocol for the spine including different qualitative as well as quantitative sequences. We obtained good correlations between muscle strength and MRI parameters (Figure 1, doi: 10.1002/jmri.26679 doi: 10.1097/RCT.0000000000000374) and were able to establish normative values for trunk muscle composition (doi: 10.3390/nu10121972 doi: 10.3389/fendo.2018.00563 doi: 10.3389/fendo.2018.00141). In our latest work, we were also able to quantify pathologic changes in nerve roots, correlating with back pain (doi: 10.1007/s10334-020-00832-w).

2. Computation and Modelling
a) Segmentation and Quantification
To overcome previous limitations in vertebra segmentation, we created and published a large-scale open-source dataset (VerSe2019: https://osf.io/nqjyw/ VerSe2020: https://osf.io/t98fz/) to serve as a reference and training dataset (https://doi.org/10.1148/ryai.2020190138). We conducted two segmentation challenges during MICCAI 2019 and 2020 with 26 participants from all over the world (https://arxiv.org/abs/2001.09193) and we launched a website to make our algorithm accessible for other researchers (https://anduin.bonescreen.de). Our own labelling and segmentation algorithm (arXiv:1902.02205 arXiv:1804.01307 arXiv:1703.04347) is on par with the top performing algorithms world-wide. From such an existing label mask we are able to extract geometrical features by automatically extracting the geometrical points of interest (GPOI, Figure 2) such as the centroid, vertebral body endplates, the spinal canal, ligament and muscle insertion points to be used for multi body simulations (not yet published). Further, we developed an algorithm to estimate a 3D model of the spine from 2D radiographs (https://arxiv.org/pdf/2007.06612.pdf). This gives us the opportunity to also use patient data for which a CT is not available for the whole spine.


b) Multi-Body Simulation (MBS) of the Spine
In order to provide boundary conditions for the investigation of loads on single spinal segments, an individualized multi-body system of the spine was developed. The current model automatically integrates patient-specific imaging data, processed with the above-mentioned pipeline, such as spinal alignment, patient's standing posture, soft tissues and body composition and material properties of bones, discs and muscles. We use the MBS-Software SIMPACK with a Matlab integration for an inverse simulation to optimize muscle forces. The current version of the model uses data from CT scans to model the posture and spinal alignment of the patient. Intervertebral discs are modelled as spring-damper systems located in each center of joint between two vertebrae. Muscular structures are modelled in a way that allows integrating their mechanical properties as derived by multiscale MRI (doi: 10.1002/jmri.26679 doi: 10.1097/RCT.0000000000000374). First simulations showed a strong influence of bodyweight, bodyweight distribution and center of gravity on muscular activation and interspinal forces. We were also able to demonstrate significant differences in load on intervertebral discs in patients with vs. without vertebral insufficiency fractures (Figure 4, paper in progress).

c) Finite Cell Method (FCM)
Based on MBS results and material properties from calibrated CT we apply the finite cell method (FCM) for the simulation of the mechanical interactions between the implanted material and the vertebrae (Figure 3) as well as the calculation of vertebral failure load (https://doi.org/10.1002/cnm.2951). First results showed significantly decreased failure loads in patients with vertebral fractures, but not in patients with back pain.

Clinical validation
In a retrospective matched-pair approach, we were able to differentiate patients with successful operation from those without (AUC = 0.77 doi: 10.3389/fendo.2020.552719). This work is currently being extended to include biomechanical analysis.
After the above-described MR- protocol was validated in a population of healthy control subjects, we started to scan patients suffering from lower back pain with and without spinal canal stenosis. We examined 40 patients suffering from lower back pain. Based on the degree and course of their pain, we found significant differences in loading of the facet joints and intervertebral discs as revealed by biomechanical modelling (paper in progress)."
1. Fully automated spinal segmentation in CT datasets is now possible due to our work: We published a large-scale open-source dataset (VerSe2019: https://osf.io/nqjyw/ VerSe2020: https://osf.io/t98fz/) to serve as a reference and training dataset (https://doi.org/10.1148/ryai.2020190138). We conducted two segmentation challenges during MICCAI 2019 and 2020 with 26 participants from all over the world (https://arxiv.org/abs/2001.09193) and we launched a website to make our algorithm accessible for other researchers (https://anduin.bonescreen.de).
2. Based our work on iterative CT reconstruction (e.g. DOI: 10.1007/s00330-017-4904-y DOI: 10.1111/jon.12810) we had a strong interaction with Philips Healthcare in this regard and consecutive developments allowed us this year to reduce the radiation dose of all major clinical protocols significantly (average ~30% reduction rate, ranging from ~25% to 75%; prospective evaluation in progress).
3. We were able to compute multi-body-simulations and finite cell models in a fully automated fashion. This approach better predicts surgical outcome than simple material properties of the spine (https://doi.org/10.3389/fendo.2020.552719) and was able to show substantial differences in spinal loading depending on a patient´s posture and on the spinal level (e.g. 45° bending vs. upright at L5: +210% load on the L5/S1 disc).
DTI of muscles, embedding of vertebrae, finite cell method of implants and multi body simulations