Technical note
Sensitivity analysis of geometric errors in additive manufacturing medical models

https://doi.org/10.1016/j.medengphy.2015.01.009Get rights and content

Highlights

  • In additive manufacturing each building step introduces errors to the models.

  • Models errors are concentrated at regions with high curvatures.

  • Triangulation resolution and segmentation threshold are critical error factors.

  • Errors could be reduced choosing better triangulation and printing resolutions.

  • More robust segmentation algorithms should be considered for the building process.

Abstract

Additive manufacturing (AM) models are used in medical applications for surgical planning, prosthesis design and teaching. For these applications, the accuracy of the AM models is essential. Unfortunately, this accuracy is compromised due to errors introduced by each of the building steps: image acquisition, segmentation, triangulation, printing and infiltration. However, the contribution of each step to the final error remains unclear.

We performed a sensitivity analysis comparing errors obtained from a reference with those obtained modifying parameters of each building step. Our analysis considered global indexes to evaluate the overall error, and local indexes to show how this error is distributed along the surface of the AM models.

Our results show that the standard building process tends to overestimate the AM models, i.e. models are larger than the original structures. They also show that the triangulation resolution and the segmentation threshold are critical factors, and that the errors are concentrated at regions with high curvatures.

Errors could be reduced choosing better triangulation and printing resolutions, but there is an important need for modifying some of the standard building processes, particularly the segmentation algorithms.

Introduction

For many years additive manufacturing (AM) models have been used in medical applications such as surgical planning, teaching aids and simulations, customized surgical implants, prosthetics and orthotics [1], [2], [3], [4], with many benefits for patients and healthcare professionals [5], [6].

Although there are different AM building technologies, most of them consist in adding material layer by layer until the desired shape is built [7], [8]. The construction of AM medical models usually consists of four steps (Fig. 1):

  • 1.

    Acquisition: The structure of interest is scanned using computed tomography (CT), magnetic resonance imaging or other three-dimensional imaging technology.

  • 2.

    Segmentation: The object of interest is segmented out from the image using any of the available segmentation algorithms. The most standard ones are intensity thresholding and region growing.

  • 3.

    Triangulation: The surface of the segmented object is approximated by a triangular mesh, which is then exported into a STereoLithography (STL) file.

  • 4.

    Printing: The STL file is loaded into a computer that drives a 3D printer. This printer builds the AM model layer by layer.

Additionally, there might be a fifth process in which the AM models are infiltrated (manually or using vacuum pumps) with different materials to give them strength or other mechanical properties. Some AM techniques do not include the third step, thus the segmented datasets are directly exported into a slice format (e.g. SLC), which effectively bypasses the triangulation step.

Each of these steps involves choosing some methodologies and parameters. In CT, users need to set the voltage and current of the X-ray tube. Thresholds or similar parameters must be defined for the segmentation process and manual editions are sometimes needed for correcting the obtained results. The size (or size range) of the triangles must be selected a priori to define the resolution of the triangulation. The layer thickness and orientation must be set at the 3D printer, and thus the in-plane and through-plane printing resolutions are defined. Finally, if AM models are infiltrated, the material needs to be chosen.

Each of the building steps introduces errors, resulting in an AM model that is not geometrically identical to the object of interest. These errors are due to the inherent nature of each process, the presence of artifacts or the definition of non-optimal parameters. Some studies have given recommendations on how to choose thresholds and how to perform the CT acquisition, so that to reduce the effects of image artifacts [9], [10]. However, it is not clear how much each building step contributes to the final error.

In medical applications, the geometric accuracy of models is very important, since it may affect the outcome of the treatment or of the chosen application. Several methodologies have been proposed to measure the accuracy of AM models. Some authors have used linear distances between anatomical landmarks to quantify geometric errors [[2], [3],[11], [12], [13], [14], [15], [16]]. Some others have used colored surface representations to show local errors of the AM models [17], [18], [19], [20]. Finally, Arrieta et al. [21] proposed local and global metrics to quantify unambiguously the geometric errors using image-processing techniques.

Few attempts have been made to quantify the error sensitivity to some of the building steps. Galeta et al. [22] measured the error sensitivity with respect to layer thickness and orientation of the printing process and three different infiltrating substances. They used linear distances between landmarks to quantify geometric errors. Fitzwater et al. [23] analyzed error variability with respect to the current of the X-ray tube, threshold values of the segmentation, and printing technology. They quantified the errors using a mixed metric of quality indexes and linear distances between landmarks computed with a coordinate measuring machine.

Our ability to locate landmarks precisely is limited, and we are prone to introduce some artificial variability into that process [2]. Additionally, measuring geometric errors with linear distances can result into an ambiguous metric that cannot always encode those errors accurately [21].

The objective of our research is to characterize and measure the contribution of each building step to the overall geometric error. In order to overcome the ambiguities associated with landmarking processes and metrics based on linear distances, we used the approach proposed by Arrieta et al. [21] to quantify the geometric errors. Thus, we could identify the most sensitive steps and parameters of the building process of AM medical models.

Section snippets

Materials and methods

AM medical models were constructed from three cadaveric phalanges (Fig. 2) obtained from the Department of Anatomy of our University, using a standard fabrication process. The cadaveric phalanges were scanned in a CT (GE HiSpeed NX I Dual) with the following parameters: helical acquisition, 80 kV, 80 mA, slice thickness 1 mm, field of view 6 cm × 6 cm and matrix resolution 512 × 512 pixels. Phalanges were segmented from the CT images with a standard software (MimicsTM 13, Materialise®, Leuven,

Results

Table 2 shows the global errors of each experiment considering A, FP and FN. Each experiment has a single parameter variation with respect to the standard case considering all construction steps (acquisition, segmentation, triangulation, printing, infiltration). As seen in Table 2, AM models built with standard parameters were overestimated with an FP rate of over 20% and with no FN.

To facilitate the analysis, we also show the average of A, FP and FN using bar plots with one standard deviation

Discussion

AM models are not identical to the original structures because of geometric distortions. Those distortions are due to: imperfections of each of the building steps or selection of non-optimal parameters within them. However, the contribution of each of these steps to the overall error remains unclear.

In order to answer this question we built reference models from three cadaveric phalanges using standard AM procedures and standard parameters. We then built different AM models after modifying one

Funding

Grant sponsor Fondecyt 1130887.

Conflict of Interest

Authors have no conflict of interest to declare.

References (37)

  • AleidW. et al.

    Development of in-house rapid manufacturing of three-dimensional models in maxillofacial surgery

    Br J Oral Maxillofac Surg

    (2010)
  • ItataniR. et al.

    Reduction in radiation and contrast medium dose via optimization of low-kilovoltage CT protocols using a hybrid iterative reconstruction algorithm at 256-slice body CT: Phantom study and clinical correlation

    Clin Radiol

    (2013)
  • HeL. et al.

    A comparative study of deformable contour methods on medical image segmentation

    Image Vis Comput

    (2008)
  • GeQ. et al.

    A robust patch-statistical active contour model for image segmentation

    Pattern Recognit Lett

    (2012)
  • NooraniR.

    Rapid prototyping: Principles and applications

    (2006)
  • BibbR.

    Medical modelling: The application of advanced design and development techniques in medicine

    (2006)
  • DuH. et al.

    Use of patient-specific templates in hip resurfacing arthroplasty: Experience from sixteen cases

    Int Orthop

    (2013)
  • ChuaC.K. et al.

    Rapid prototyping: Principles and applications

    (2010)
  • Cited by (0)

    View full text