A NEW METHOD OF 3D PREOPERATIVE TKA SIZING AND PLACEMENT FROM 2D ANALYSIS OF
CALIBRATED RADIOGRAPHS Jerome Grondin Lazazzera, Derek Cooke, Mike Brean, James Stewart
Queens Univ, Computing Sciences, OAISYS Inc.
Summary
Our aim is to generate a patient-specific surface mesh (PSSM) preoperatively from calibrated planar
knee radiographs and a Statistical Shape Atlas (SSA) used to determine implant sizing and rotational
alignment in TKA. Anatomical features were annotated on the 2D calibrated images and their threedimensional positions recovered using a perspective-n-point algorithm. The SSA was derived from
segmented Computed Tomography (CT) images of the knee and embedded in the triangular surface
meshes; key features were annotated with a reliable method for spatial location using a sagittal plane
for reference and embedded in the surface meshes. The radiographic feature annotations will be used
to guide deformation of the mean shape into a PSSM. A reliable method for preoperative implant sizing
and fit using planar radiographs has potential to cut costs and improve TKA outcomes.
Objectives:
To generate a patient-specific surface mesh (PSSM) preoperatively, which is used to determine implant
sizing and rotational alignment in TKA from calibrated planar knee radiographs and a Statistical Shape
Atlas (SSA). These approaches are aimed to compete with existing methodologies relying on 3D imaging.
Materials and Methods:
The planar radiographs required means for landmark identification and calibration. Corresponding
anatomical features are annotated on the 2D images using a calibrated frame and their threedimensional positions recovered using a perspective-n-point algorithm. These annotations are used to
generate a PSSM from the SSA. Constructing the SSA entailed multiple steps: segmenting Computed
Tomography (CT) images of the knee to derive triangular surface meshes; selection of key anatomical
features with a reliable method for spatial location using a sagittal plane for reference; embedding
features, annotated on the CT images, into the surface meshes. The SSA is created when all the meshes
are deformed into a mean shape. Finally, radiograph feature annotations are used to guide deformation
of the mean shape into a PSSM.
Results:
We have created a mobile positional frame, featuring fiducial markers in radiolucent walls, along with
the software needed to recover 3D dimensionality from calibrated radiographs. CT scans of 16 right and
15 left femurs were manually segmented and triangular meshes were derived using a curvaturedependent implicit surface extraction approach. A set of key anatomical features was established based
on literary research and expert surgical consultation. Additionally, a femoral sagittal plane consisting of
the anatomical axis and the posterior femoral point was defined as a unique reference to improve
consistence of feature identification. Software was built to define and collect the key landmarks.
Annotations of all images were completed by a surgeon and two trained readers. Inter-observer
variability analysis was conducted to assess feature reliability. The annotations, embedded into the
mesh models, were used as initialization for rigid registration using an iterative closest point algorithm.
Non-rigid registration was performed using a thin plate spline point matching algorithm. Principal
component analysis was used to construct the statistical shape atlas and the leave-one-out method was
used to determine the reconstruction accuracy. Results from both feature variability analysis and
reconstruction testing showed promise. Additionally, the method has provided novel information on
femoral dimensions which includes sizing of medial & lateral condylar depths, femoral widths, their
aspect ratios and condylar radii. New geometric data include distal femoral condylar tangents to the
anatomic axes and transcondylar alignment to the anatomical axis.
Conclusion:
A varied selection of segmented CT image sets were analyzed using a sagittal plane based reference
approach for the annotation of key landmarks. This method of 3D analysis is reproducible and has
provided new information on knee geometry with good inter-observer reliability. Reconstruction results
using 2D renderings from CT images, based on the leave-one-out method, showed promise and viability
of our method. Next steps include matching implant dimensional data to SSA; clinical studies of
matched CT and 2D imaged cases. Significance: Use of 2D radiographs instead of 3D imaging
preoperatively reduces implant inventory, improves access, functional outcomes and reduces costs.
Abbreviated Conclusion for web intro
“We have demonstrated that a method of 3D model reconstruction from annotated 2D images obtained
within a reference frame accurately reproduces knee geometry with good inter-observer reliability. The
method may be used instead of 3D imaging before and in surgery for TKA sizing and placement with
widespread access for use, improved outcomes for much less cost”.