ASTM INTERNATIONAL

Helping our world work better

Additive
Manufacturing
for Medical Data:
Optimized Medical
Image Data

ISO/ASTMTR52916
Additive Manufacturing for
Medical Data: Optimized
Medical Image Data
ASTM STOCK NUMBER: ISO/ASTMTR52916-EB
DOI: 10.1520/ISO/ASTMTR52916-EB

ASTM International
100 Barr Harbor Drive
PO Box C700
West Conshohocken, PA 19428-2959

ISO Copyright Office

Ch. De Blandonnet 8 - CP401
CH-1214 Vernier, Geneva, Switzerland
ISBN-EB: 978-0-8031-7154-1

Copyright V
C 2022 ASTM International. All rights reserved. This material may not be reproduced or copied, in whole or in part, in any printed,

mechanical, electronic, film, or other distribution and storage media, without the written consent of the publisher.

ASTM Photocopy Rights Authorization to photocopy items for internal, personal, or educational classroom use, or the internal, personal, or
educational classroom use of specific clients, is granted by ASTM International provided that the appropriate fee is paid to the Copyright
Clearance Center, 222 Rosewood Drive, Danvers, MA 01923, Tel: (978) 646-2600; http://www.copyright.com/

Co-Publisher:
ASTM International
100 Barr Harbor Drive
PO Box C700
West Conshohocken, PA 19428-2959
Phone: (610) 832-9585
Fax: (610) 832-9555
ISBN-EB: 978-0-8031-7154-1
ASTM Stock Number: ISO/ASTMTR52916-EB
DOI: 10.1520/ISO/ASTMTR52916-EB

Cite as ASTM International, Additive Manufacturing for Medical Data: Optimized Medical Image Data (West Conshohocken, PA: ASTM
International, 2022), https://doi.org/10.1520/ISO/ASTMTR52916-EB

ASTM International is not responsible, as a body, for the statements and opinions expressed in this publication. ASTM International does not
endorse any products represented in this publication.

Copyright by ASTM Int’l (all rights reserved); Downloaded/printed by ASTM International (ASTM International) pursuant to License Agreement.
No further reproductions authorized.

Co-Publisher:
ISO copyright office
Ch. De Blandonnet 8 – CP401
CH-1214 Vernier, Geneva, Switzerland
Tel. +41 22 749 01 11
Fax +41 22 749 09 47
copyright@iso.org
www.iso.org
ISO/ASTM 52916:2022€
ISO/TE 261/SC/WG
Secretariat: DIN

Additive Manufacturing for Medical Data: Optimized Medical Image Data

Fabrication additive dans le secteur médical Données: Données d’images médicales optimisées
 iii

Contents

Foreword v
Introduction vi
1 Scope 1
2 Normative References 1
3 Terms and Definitions 1
3.1 CT 2
3.2 MRI 2
3.3 Polygon 2
3.4 Reconstruction 2
3.5 Rendering 2
3.6 ROI 2
3.7 Segmentation 2
3.8 Volume data 2
3.9 VOXEL 2
3.10 2D 2
3.11 DICOM 2
4 Medical Images Generation for AM 3
4.1 General Medical Image Data Generation 3
4.2 General Error Occurrence Steps in Medical Images Generation 3
4.3 Medical Image Extraction 4
4.3.1 Introduction of medical image extraction 4
4.3.2 CT image error generation factors 4
4.3.3 MRI Image error generation factors 4
5 Image Segmentation 6
5.1 Introduction of Segmentation 6
5.2 Segmentation Techniques 6
5.2.1 Thresholding algorithm 6
5.2.2 Region growing algorithm 6
5.2.3 Morphological image algorithm 6
5.2.4 Level-set algorithm 7
5.2.5 Other partial segmentation algorithms 7
6 Reconstruction 7
6.1 Introduction of Reconstruction 7
6.2 Reconstruction Process 7
7 Smoothing 7
7.1 Marching Cubes 7
7.2 Mesh Smoothing 7
iv CONTENTS

8 3D Visualization Method 8
8.1 Surface Rendering 8
8.1.1 Introduction of surface-shaded rendering 8
8.1.2 Surface-shaded rendering feature 8
8.2 Volume Rendering 8
8.2.1 Introduction to volume rendering 8
8.2.2 Volume rendering features 9
8.2.3 Ray casting techniques 9
8.2.4 3D texture mapping techniques 9
9 Additional AM Processing 9
10 Methods 9
10.1 Image Isotropic Conversion 9
10.2 Image Enhancement 11
10.3 Image Segmentation 11
11 Minimizing Software and Equipment Errors 13
11.1 Introduction to Software and Equipment Errors 13
11.2 Software Errors 13
11.2.1 Background 13
11.2.2 Verification method using main inflection 13
11.2.3 Improving accuracy and precision 13
11.3 Equipment errors 14
11.3.1 Background 14
11.3.2 Standard computational mesh model data creation for an evaluation method 14
11.4 Tolerance Error Situations 14
Appendix A (Informative): Medical CAD for Additive Manufacturing Tolerance 14
A.1 Purpose 14
A.2 Procedure 15
A.3 Results 20
References 20
 v

Foreword

ISO (the International Organization for Standardization) is a worldwide federation of national standards bodies
(ISO member bodies). The work of preparing International Standards normally is carried out by ISO technical
committees. Each member body interested in a subject for which a technical committee has been established
has the right to be represented on that committee. International organizations, governmental and nongovern-
mental, in liaison with ISO, also take part in the work. ISO collaborates closely with the International Electro-
technical Commission (IEC) on all matters of electrotechnical standardization.
The procedures used to develop this document and those intended for its further maintenance are
described in the ISO/IEC Directives, Part 1. In particular, the different approval criteria needed for the different
types of ISO documents should be noted. This document was drafted in accordance with the editorial rules of
the ISO/IEC Directives, Part 2 (see www.iso.org/directives).
Please note the possibility that some of the elements of this document may be the subject of patent rights.
ISO shall not be held responsible for identifying any or all such patent rights. Details of any patent rights identi-
fied during the development of the document will be noted in the Introduction or on the ISO list of patent dec-
larations received (see www.iso.org/patents).
Any trade name used in this document is information given for the convenience of users and does not con-
stitute an endorsement.
For an explanation of the voluntary nature of standards, the meaning of ISO-specific terms and expressions
related to conformity assessment, as well as information about ISO’s adherence to the World Trade Organiza-
tion (WTO) principles in the Technical Barriers to Trade (TBT), see www.iso.org/iso/foreword.html.
This document was prepared by ISO/TC 261, Additive Manufacturing, in cooperation with ASTM Commit-
tee F42 on Additive Manufacturing Technologies on the basis of a partnership agreement between ISO and
ASTM International with the aim to create a common set of ISO/ASTM standards on additive manufacturing,
and in collaboration with the European Committee for Standardization (CEN) Technical Committee CEN/TC
438, Additive Manufacturing, in accordance with the Agreement on Technical Cooperation between ISO and
CEN (Vienna Agreement).
Any feedback or questions on this document should be directed to the user’s national standards body.
A complete listing of these bodies can be found at www.iso.org/members.html.
vi

Introduction

This document has been developed in cooperation with ISO/TC 261 and ASTM Committee F42 on the basis of
a partnership agreement between ISO and ASTM International with the aim to create a common set of ISO/
ASTM standards on additive manufacturing (AM).
Digital imaging and communications in medicine (DICOM) image files cannot be used directly for three-
dimensional (3D) printing; further steps are necessary to make these files readable by the AM system. In partic-
ular, as the thickness of the computed tomography slice increases, problems associated with errors in 3D
reconstruction of the anatomical structure also increase. Therefore, the focus of this technical report is to auto-
matically reconfigure the slice interval through the application of isotropic conversion technology to utilize the
existing DICOM file and visualization and editing software. In addition, to present a method for optimized
medical image data for AM, tomography metadata without compression is used by editing and processing the
output format file without loss in the AM equipment system, or tomography within the maximum allowable
range of radiation. Consider reducing the spacing of slices as much as possible and increasing the resolution per
image as much as possible.
This document benefits from the direction of development and high-quality AM output through the techni-
cal optimization of medical imaging for AM, including medical academics, clinic and industry fields for AM
such as anatomical measurements, 3D analysis, finite element analysis and surgical planning or simulation,
patient-specific implant, and device design. Affected stakeholders include medical AM system manufacturers,
AM feedstock manufacturers, AM feedstock suppliers and vendors, medical AM hardware manufacturers, medi-
cal AM software manufacturers, medical AM system manufacturers, medical AM platform manufacturers,
AM-based medical device manufacturers, medical 3D scanning and digitizing device manufacturers, surgical
simulation AM model manufacturers, AM surgical implant manufacturers, AM surgical guide manufacturers,
AM physical models for clinical education and diagnostic treatment, and disposable medical AM consumable
devices.
doi:10.1520/ISO/ASTMTR52916-EB/ available online at www.astm.org

Additive Manufacturing for Medical

Data: Optimized Medical Image Data

Reference
ASTM International, Additive Manufacturing for Medical Data: Optimized Medical Image Data (West
Conshohocken, PA: ASTM International, 2022), https://doi.org/10.1520/ISO/ASTMTR52916-EB

1 Scope
This document includes the creation of optimized data for medical additive
manufacturing (MAM). These data are generated from static modalities, such as
.
magnetic resonance imaging (MRI) and computed tomography (CT). This docu-
ment addresses improved medical image data, and medical image data acquisition
processing, and optimization approaches for accurate solid medical models, based
on real human and animal data.
Solid medical models generally are created from stacked two-dimensional (2D)
images output from medical imaging systems. The accuracy of the final model
depends on the resolution and accuracy of the original image data. The main factors
influencing accuracy are the resolution of the image, the amount of image noise, the
contrast between the tissues of interest, and artifacts inherent in the imaging system.

2 Normative References
The following documents are referred to in the text in such a way that some or all
of their content constitutes the requirements of this document. For dated references,
only the edition cited applies. For undated references, the latest edition of the refer-
enced document (including any amendments) applies.
ISO/ASTM 52900, Additive Manufacturing—General Principles—Fundamentals
and Vocabulary

3 Terms and Definitions

For the purposes of this document, the terms and definitions given in ISO/ASTM
52900 and the following apply.

Copyright V
C 2022 by ASTM International, 100 Barr Harbor Dr., PO Box C700, West Conshohocken, PA 19428-2959 1
2 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

ISO and IEC maintain terminology databases for use in standardization at the following addresses:
• ISO Online Browsing Platform: https://www.iso.org/obp
• IEC Electropedia: https://www.electropedia.org
3.1 CT
computed tomography
computed axial tomography
radiographic scanning technique that uses a number of CT projections of an object at different angles in order
to allow calculation of a CT image (see ISO 15708-1:2017, 3.7)
3.2 MRI
magnetic resonance image
imaging technique that uses static and time varying magnetic fields to provide images of tissue by the magnetic
resonance of nuclei (see ISO 14630:2012, 3.5)
3.3 POLYGON
planar surface defined by one exterior boundary and by zero or more interior boundaries (see ISO 11783-
10:2015, 3.13)

Note: Each interior boundary describes a hole in the surface. A single or group of polygons can be used to define
a treatment zone.

3.4 RECONSTRUCTION
process of transforming a set of CT projections into a CT image (see ISO 15708-1:2017, 3.25)
3.5 RENDERING
action of transforming from a scene description to a specific output description/device (see ISO 19262:2015, 3.213)
3.6 ROI
region of interest, sub-volume within an object or a CT image (see ISO 15708-1:2017, 3.26)
3.7 SEGMENTATION
method which partitions a surface or volume into distinct regions (see ISO 25178 2:2012, 3.3.6, modified; ISO
25178 2:2012 had “scale-limited surface” in the definition)
3.8 VOLUME DATA
data of a volume in a 3D space (see ISO 18739:2016, 3.1.42)

Note: The description can be performed on the basis of density differences inside the three-dimensional space.
3.9 VOXEL
volume pixel
three-dimensional cuboid representing the minimum unit comprising a three-dimensional image (see ISO/TR
16379:2014, 2.17, modified; “volume pixel” has been added as a second term)
3.10 2D
geometry in a xy-plane, where all the geometry’s points have only x and y coordinates (see ISO 14649-10:2004, 3.1)
3.11 DICOM
digital imaging and communications in medicine
international standard for medical images and related information

Note: It defines the formats for medical images that can be exchanged with the data and quality necessary for
clinical use. The Medical Imaging Technology Association (MITA), a division of NEMA, serves as the DICOM
Secretariat. The current DICOM standard may be found at https://www.dicomstandard.org/current.
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 3

4 Medical Images Generation for AM

4.1 GENERAL MEDICAL IMAGE DATA GENERATION
The start for image generation is to collect raw image data. This collects raw information about the inside of the
human body and becomes the basic object of all subsequent image processing tasks. In the end, regardless of the
image format, the data collection process detects physical factors, preprocesses the collected signals, and then
digitizes them (see fig. 1).1

4.2 GENERAL ERROR OCCURRENCE STEPS IN MEDICAL IMAGES GENERATION

With gradual technological advancement, many solutions for medical AM are emerging. Research into the cause
for resolving errors in medical AM output is still ongoing. The cause of AM accuracy error occurs in the process
of converting the raw data to medical images and the process of converting 3D model data.2 Error generation
factors that occur during this conversion process are described in section 4.3 for the most common tomography
systems.
Additional errors may be generated by the process of converting DICOM or PACS data to the computa-
tional formats used within segmentation editing software and saving the STL 3D mesh format for use in AM
systems. When saving a customized STL file, all meta data that defined color, material, and surface textures are
lost. The lack of accuracy and precision for 3D data from the scan systems, editing and modeling software can
reduce the quality of an AM medical device.
Note 1: There can be other factors in creating errors when utilizing other image capture modalities, such as
ultrasound and digital microscopy, not covered in section 4.3.

FIG. 1 Process from medical image to medical additive manufacturing.

4 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

4.3 MEDICAL IMAGE EXTRACTION

4.3.1 Introduction of medical image extraction
The quality of a medical image depends on the degree to which the microscopic structure of the human body
can be accurately represented. According to the needs of the medical professional who requested the tomogra-
phy, the layer spacing between the cross-sectional images is adjusted and photographed. Based on the captured
meta data, reconstruction through 3D visualization is performed to extract the data of the region of interest. In
this process, the medical imaging tomography technology, imaging conditions, and data conversion process will
continue to affect the medical AM output resolution.3

4.3.2 CT image error generation factors

CT modality images use absorption coefficient parameters that visualize the density of an image. The contrast of
hard tissue is more clearly expressed than soft tissue. Because sequential image layers are output as a series, 3D
reconstruction is possible. The important factors that determine the image quality are the accuracy of the CT
reduction coefficient, which expresses the degree of attenuation of a substance, noise, uniformity, spatial resolu-
tion, contrast resolution, and radiation dose. It is recommended that the patient’s exposure dose is small, but it
is very difficult to control the exposure dose and image quality because it is directly related to image noise and
density resolution. Adjustment of radiation dose for each body part according to the patient’s condition follows
the clinical experience and medical recommendations of the radiologist. This is an external factor that affects
the medical image data homogeneity.
• CT matrix size: The digital medical image is stored as 2D pixels, and each pixel is converted into the
number of bits matched by the number of gray levels and represented. The CT image size depends on
the anatomy being examined. Typically, CT images have a matrix size of 512 pixels  512 pixels  8 bytes
(12 bits), and gray levels range from 512 pixels (28 bits) to 4 096 pixels (212 bits). A single CT section
requires 512 pixels  512 pixels  2 bytes = 524,288 bytes of storage on the computer.
• CT reduction coefficient: The tissue weighting factor (WT) is a relative measure of the risk of stochastic
effects that might result from irradiation of that specific tissue. It accounts for the variable radio-
sensitivities of organs and tissues in the body to ionizing radiation. To calculate the effective dose, the
individual organ equivalent dose values are multiplied by the respective tissue weighting factor and the
products added. The sum of the weighting factors is 1.
• Based on the values of tissue weighting factors, tissues are grouped into following to assess the carcino-
genic risk:
• high risk (WT = 0,12): stomach, colon, lung, and red bone marrow;
• moderate risk (WT = 0,05): urinary bladder, esophagus, breast, liver, and thyroid; and
• low risk (WT = 0,01): bone surface and skin.
• Spatial resolution: Ability to image small objects that have high subject contrast; CT has moderate spatial
resolution 20 lp/cm.
• Contrast resolution: Ability to distinguish between and image similar tissues; CT has excellent low-
contrast detectability 0.25% to 0.5% difference in tissue attenuation.

4.3.3 MRI image error generation factors

MRI uses a magnetic field that is harmless to the human body and radio frequency, which is nonionizing radia-
tion. The principle is to image the density and physicochemical properties of the atomic nucleus by causing
nuclear magnetic resonance phenomenon in the atomic nucleus inside the human body. The advantage and dif-
ference are that it has several imaging parameters compared with CT. The density of the hydrogen atom nucleus,
T1 relaxation time, T2 relaxation time, and blood flow are the four important parameters that determine the
shading of the image. Not only the distribution of the hydrogen atom nucleus varies but also the molecular state
of the contained tissue or the physical state of the image. The MRI image looks at the distribution of spin density
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 5

and is further affected by the T1-T2 relaxation time associated with the nuclear magnetic resonance (NMR) phe-
nomenon. Because of the parameter elements for each MRI device, standard parameter settings are different for
each MRI imaging personnel, and these external factors affect image data homogeneity.
MRI image quality depends on resolution (e.g., matrix, field of view, slice thickness), signal-to-noise ratio,
contrast, and artifacts. In particular, contrast depends on the MRI scan parameter.
MRI resolution is the size of an individual pixel—the smaller it is, the higher the resolution. The MRI
matrix size is the number of pixels in the images. To improve the MRI resolution, increase the matrix, decrease
the field of view (FOV), and decrease the slice thickness.
In the field of orthopedic surgery, MRI scan parameters are applied in the following ranges of maximum
and minimum values of FOV, slice thickness, interslice gap, and matrix size (see table 1).
• Signal-to-noise ratio: The signal-to-noise ratio is a measure that compares the level of a desired signal to the
level of background noise. For data acquired through magnetic resonance imaging, this quantification typically
is used to allow for a comparison between imaging hardware, imaging protocols, and acquisition sequences. In
this context, the signal-to-noise ratio is conceptualized by comparing the signal of the MRI image to the back-
ground noise of the image. Mathematically, the signal-to-noise ratio is the quotient of the signal intensity mea-
sured in a region of interest and the standard deviation of the signal intensity in a region outside the anatomy
of the object being imaged or the standard deviation from the noise distribution when known. For example,
FOV, scan parameters, magnetic field strength and slice thickness, and the signal-to-noise ratio of MRI images
can be increased because these parameters influence the background noise.
• Image contrast: The repetition time (TR) and echo time (TE) are basic pulse sequence parameters. They
typically are measured in milliseconds (ms). The TE represents the time from the center of the RF-pulse
to the center of the echo. For pulse sequences with multiple echoes between each RF pulse, several echo
times may be defined and are commonly noted TE1, TE2, TE3, and so on. TR is the length of time
between corresponding consecutive points on a repeating series of pulses and echoes.

Variations in the value of TR and TE have an important effect on the control of image contrast characteris-
tics. Short values of TR (e.g., less than 1,000 ms) and TE (e.g., less than 25 ms) are common in images exhibiting
T1 contrast. Long values of TR (e.g., 1,500 ms) and TE (e.g., greater than 60 ms) are common in images exhibit-
ing T2 contrast. Middle TR values (e.g., from 1,000 ms to 1,500 ms) and middle TE values (e.g., from 25 ms to
60 ms) are common for density-weighted contrast.
• Artifacts: Most common MRI artifacts were movement, Gibbs, metal and slice overlap artifacts.
• Movement artifact correction: Breath holding, sedation, anesthesia, electrocardiographic trigger, spatial
RF presaturation, and flow compensation.
• Gibbs artifact (truncation, ringing, spectral leakage artifacts) correction: Softening filters, larger acquisi-
tion matrix, and smaller FOV.
• Slice overlap artifact correction: Avoid sharp angle changes between slice groups, increase a gap between
slices, and apply different storage processes for images.

TABLE 1 Musculoskeletal MRI scan parameters

Scan section

Parameters Shoulder Elbow Wrist Hip Knee Ankle

Field of view (cm) 16 10 to 16 6 to 12 16 to 20 16 14
Slice thickness (mm) 3 3 to 4 3 3 to 4 3 3
Slice gap (%) 10 33 33 33 10 10
Matrix size (pixel)  256  192  256  256  256  192  512  384  256  192  256  192
6 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

5 Image Segmentation
5.1 INTRODUCTION OF SEGMENTATION
Image segmentation is the process of partitioning an image into multiple labeled regions locating objects and
boundaries in images. It can be used to create patient-specific, highly accurate computer models of organs and
tissue. There are a number of image segmentation techniques, and each has advantages and disadvantages, but
no single segmentation technique is suitable for all images and applications. Basic segmentation approaches rely
on the principle that each tissue type has a characteristic range of pixel intensities. Hence, it is possible to distin-
guish between tissues and identify boundaries.3
Image segmentation refers to a process of grouping connected pixels having similar characteristics among
pixels constituting a given whole image. Despite the many image processing methods, however, there are not
many differences in the image attribute information values of the anatomical human structures. Thus, many
results appear to be unclear or disconnected. If a modification is made to improve image quality, data corruption
problems occur. In the end, it is necessary to improve the segmentation algorithm that can extract all region of
interest (ROI) boundaries.3,9

5.2 SEGMENTATION TECHNIQUES

In the 2D medical image, the region of interest needs to be accurately divided so that the desired region of inter-
est can be 3D visualized. As a measure of image segmentation grouping, feature elements such as contrast, color
components, edges, texture, motion, and depth information are used. Many types of segmentation algorithms
are applied based on these indicators, and the image segmentation methods. Thresholding-based algorithms,
clustering-based algorithms, region-based algorithms, and level-set-based algorithms are representative.4

5.2.1 Thresholding algorithm

Thresholding algorithm is a method to divide into a thresholding range using a histogram. In this case, when a
characteristic of a pixel is a pixel value, a set of pixels with a result of 1 is called an object, and a set of results
with a result of 0 is called a binary image partitioning method. After determining the general threshold values
for bones and muscles, segments of all pixels larger or smaller are divided into groups, and segments are sequen-
tially processed. If, however, there is no spatial characteristic of the image to make the noise stronger, there is a
method of segmentation using information associated with the local intensity.

5.2.2 Region growing algorithm

Region growing segmentation is a method of finally dividing the entire image by gradually integrating and growing
regions with the same characteristics from adjacent small regions. This algorithm regards the starting point of region
growth as a seed, and if the similarity is greater than a certain threshold, it is regarded as the same region. In the case
of an image in which adjacent pixel values change little continuously, an incorrect segmentation performed. There-
fore, by using the average feature of the entire integrated area to date, a boundary area larger than a threshold value
can be determined by comparing it with the information of the next integrated object pixel.

5.2.3 Morphological image algorithm

In image analysis, expressing the region’s shape uses the concept of mathematical morphology as a tool to extract
useful image elements. The language of mathematical morphology is set theory, and it provides an intensive and use-
ful method for image-processing problems. Frequently used logical operations are AND, OR, and NOT, which are
executed between pixels based on the corresponding pixels of the image. Useful methods in binary morphology are
dilation and erosion opening and closing. The application stage serves to extract image components useful for expres-
sion and description of the form. The image component extraction step for application is used along with the mor-
phological algorithm to extract the boundaries, the connected elements, and the skeleton of the region—including,
for example, boundary extraction, area filling, and connection element extraction.
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 7

5.2.4 Level-set algorithm

Level-set processing is a segmentation method that creates contour lines by connecting pixels of the same char-
acteristic in a 2D image surface with a curve and maintains the natural shape through grouping and connection
algorithms of the same area. This method distinguishes more accurate pixel values and enables easy division by
implementing grouping easily. This can be implemented easily and smoothly when the polygon is applied to a
3D model through an extended application.

5.2.5 Other partial segmentation algorithms

Another segmentation method directly displays an area (clustering) on the input image. A second method
moves to a specific space, divides, and then extracts back to the original image. A third method subdivides a
region in which features are not uniform, using the entire image as a starting point and finally stopping the divi-
sion in a region in which all pixels are uniform. A fourth method defines a uniform region after extracting a
closed curve using only the extracted edge information.

6 Reconstruction
6.1 INTRODUCTION OF RECONSTRUCTION
It is difficult to distinguish the anatomically correct structure or location of a lesion inside the human body by
using only the basic data obtained by the medical tomography equipment and the 2D cross-sectional image.
Therefore, it is necessary to reconstruct in three dimensions to accurately grasp the location and structure of the
lesion, the structure and size of each organ, and the treatment site.

6.2 RECONSTRUCTION PROCESS

Reconstruction is a series of processes used to scan the 2D medical image through CT MRI equipment to check the
lesions in the human body and visualize and extract the segmented data of the obtained 2D medical image series in a
3D shape. It is defined as a process of removing the cause affecting the sample shape measurement result from the
image and obtaining an estimate of the measured sample surface topography. Specifically, it is composed by filling
the 3D surface with triangular polygons according to the phase difference, when the outline of the section of the visu-
ally divided region is connected in the vertical direction, and the quality of the digital shape is determined. This
implies that it is necessary to perform a proper division process in multiple steps in the previous step, and the visuali-
zation output process for confirming the result also must be performed multiple times. It has to be distinguished,
however, from image reconstruction in 2D medical imaging and 3D reconstruction of 3D models.

7 Smoothing
7.1 MARCHING CUBES
The “marching cubes” is a simple iterative algorithm for creating triangular surfaces for a 3D function (in our
case, the 3D function is defined point wise and is called voxels). It works by “marching” over the whole 3D
region, which has been divided into cubes. The vertices of the cube are the voxels.
The algorithm then computes whether a triangular surface passes through this cube or not. Through the
marching cubes algorithm, the extracted region of the medical image consisting of voxels can be converted into
a mesh-type data structure.

7.2 MESH SMOOTHING

In general, because the image was converted to a mesh through DICOM voxel, it basically maintains the shape
of a hexahedron. Therefore, it is necessary to apply a method of representing a smooth surface through the spec-
ification of a coarse polygon mesh. Smooth surfaces can be computed from coarse meshes as a limitation of the
iterative process of subdividing each polygon face into smaller faces closer to the smooth surface (see fig. 2).
8 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

FIG. 2 Reconstruction (A) before and (B) after by mesh smoothing.

8 3D Visualization Method
8.1 SURFACE RENDERING
8.1.1 Introduction of surface-shaded rendering
The surface-shaded rendering technique visually reconstructs the perspective of the anatomical elevation of a
shaded object through the reflection of light from the model surface, based on shape, position, light source, and
viewpoint information. By analyzing a combination of Hounsfield ranges and segmentation techniques, it calcu-
lates the location of surfaces separating anatomic tissue types. The surface information is then used to calculate
a perspective visualization based on selectable observer position and light source positioning. With the develop-
ment of virtual lighting and shading technology, it has been possible to create realistic anatomic images in
real time.

8.1.2 Surface-shaded rendering feature

Because specific scalar values are displayed as basic shapes on the volume data and rendering is partially applied
to the generated surface, the amount of data to be processed is not large and the speed of 3D reconstruction pro-
ceeds rapidly. In displaying useful information of the entire cross-sectional image data, however, surface render-
ing limits only specific scalar values to basic shapes. Therefore, the external surface information is implemented,
but it causes loss of information that cannot simultaneously express the independent shape of the interior,
resulting in a reduction in the overall volume.

8.2 VOLUME RENDERING

8.2.1 Introduction to volume rendering
Volume rendering is a visualization technique that creates an internal image of a medical tomography image
using a volume data set, which is a set of 3D scalar intensity based on voxels. Voxels are small cuboid shapes in
a three-dimensional space and have a color density in that space. The volume information is a 3D matrix struc-
ture of these voxels. Data obtained by CT, MRI, CAD (computer-aided design), CAM (computer-aided
manufacturing), and simulation can be transformed into volume information.
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 9

8.2.2 Volume rendering features

Volume rendering goes through the preprocessing stage of data extraction, gradient application, resampling to
perform 3D interpolation, classification to emphasize the boundary by applying opacity, and sequential synthesis
process to combine numerous samples that assign color to data into one sample. In this compositing method,
when a ray passes through each pixel and samples the pixel values and positions, a ray casting method creates
an image order that generates images by synthesizing the opacity and intensity of voxels in which each ray lies
in a straight line. In addition, using an object order method, voxels are searched in the order in which the vol-
ume data are stored and synthesized into pixels corresponding to each voxel. The basic unit on the 3D image,
the voxel includes coordinate information on the X, Y, Z axis, color value information, density value, and opac-
ity information.5 By assigning opacity and shading to each voxel, and by projecting and compositing these
values, the samples are expressed in a three-dimensional image plane.

8.2.3 Ray casting techniques

Ray casting is a technique in which a virtual ray is shot at each pixel of the viewpoint plane in the space where
volume information is located. This technique blends voxels that the ray meets, which are stored in the pixel
where the ray is started to generate the entire image. Because all of the voxels are compared for each ray to deter-
mine which voxels the ray meets, the amount of work to be performed is quote large. Additionally, the larger
the resolution of the image to be generated, the more the amount of work is rapidly increased.

8.2.4 3D texture mapping techniques

This mapping technique applies a color value corresponding to a 3D texture to a polygon in the space when tex-
ture information about the 3D space is given. It uses a 3D texture mapping unit built into the hardware. This
unit creates a section at regular intervals in the space where the volume information is located. After mapping
the volume information to the section in 3D texture, it is possible to obtain the final alpha-blended image for
the entire sections.
The general implementation steps of 3D texture mapping are as follows: First, the volume information is
stored in the graphic memory as 3D texture data, and cross sections parallel to the viewpoint plane are generated
at regular intervals in the space where the volume information will be located. Second, the generated cross sec-
tion is mapped to a 3D texture stored in memory through a mapping unit. Third, the alpha-blended cross sec-
tions of the mapped sections are stored in the frame buffer and the contents of the buffer are displayed on the
screen. This method has disadvantages, however, including that the size of the physical memory required and
the bus bandwidth inside the graphics hardware are very large, and it is also difficult to apply a real-time shading
and early ray termination, anti-aliasing performance improvement technique.

9 Additional AM Processing
To maximize the merits of MAM, it is necessary to convert, edit, and process visualization data. For patient-specific
AM prosthetics, it is essential to design a 3D shape based on visual data inspection and consultation by medical
experts. During the entire process of generating medical image-based data, data corruption is inevitable in the edit-
ing process. Data corruption is due to the gap between the technician’s clinical anatomy knowledge and the medical
professional’s data transformation knowledge. Data transformation can continue to provide clinical anatomy data
and data transformation optimal algorithms to examine how to solve these problems with machine learning.

10 Methods
10.1 IMAGE ISOTROPIC CONVERSION
Theoretically, getting more CT images enhances the quality of reconstruction because CT data represent informa-
tion in the human anatomy. But, getting more data from a CT scan is impractical due to the fact that the human
10 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

body receives a dose of X-ray radiation. Therefore, in most cases, CT images have different depth dimensions. This
problem, called anisotropic, causes the voxel size to not be in a cubic shape, which causes the following:
• stair-step artifact reconstruction,
• wrinkled surface reconstruction, and
• degradation of original anatomy size.

Isotropic conversion is the method used to normalize the CT volume voxel size. To make the same image
spacing of CT data, this method basically uses a sampling method, by adopting a specific interpolation method,
such that the resulting volume has isotropic volume. In other words, this method adds more data in between
each slice such that the whole spacing size is the same. There are several interpolation algorithms for isotropic
conversion, such as one-dimensional nearest neighbor, linear, cubic, 2D nearest neighbor, b-spline, bilinear, and
bicubic interpolation. B-spline interpolation provides more accurate interpolation in CT cases than the other
interpolation methods.6
Figures 3 and 4 show some examples of isotropic conversion. In figure 3, the left image is anisotropic
CT data. After isotropic conversion, as shown on the right of figure 4, the resolution of the image is increased,

FIG. 3 Reconstruction (A) before and (B) after 3D medical data generated by isotropic conversion.

Resolution: 512 pixels  512 pixels  94 images Resolution: 512 pixels  512 pixels  651 images
Spacing: 0.7 mm  0.7 mm  5.0 mm Spacing: 0.7 mm  0.7 mm  0.7 mm

FIG. 4 Isotropic 3D modeling results (A) before and (B) after isotropic conversion.

Modality: MR
Resolution: 320 pixels  320 pixels  94 images
Spacing: 1.19 mm  1.19 mm  3.0 mm
Interpolation Method: None

Modality: MR
Resolution: 320 pixels  320 pixels  141 images
Spacing: 1.19 mm  1.19 mm  1.19 mm
Interpolation Method: b-spline
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 11

resulting in smooth reconstruction from a coronal view. In figure 4, it is possible to see that isotropic conversion
also increases the surface smoothness of the 3D modeling because the isotropic data depth is larger than the aniso-
tropic data depth.

10.2 IMAGE ENHANCEMENT

In medical imaging tomography, noise is inevitable because of different imaging principles. Therefore, it is nec-
essary to remove this noise before segmentation to ensure a quality segmentation result.
A Gaussian low-pass filter uses a normal distribution function to remove noise from the image (see fig. 5),
while the Laplacian of Gaussian filter is a filtering method that highlights the edges of the images (see fig. 6). By
highlighting the edges, you can either remove the noise in the medical image or preprocess the techniques that
emphasize certain areas to increase the quality of the final segmentation result.7

10.3 IMAGE SEGMENTATION

Image segmentation is the process of partitioning a digital image into multiple segments (see figs. 7 and 8). The
goal of medical image segmentation is to simplify and change the representation of an image into something
that is more meaningful and easier to analyze.
Extraction results inevitably include false positives (FP). Removal of these FP is necessary, which may
require different methods depending on the extraction algorithm.
For instance, in figure 9A, the black area is the FP area of the extraction of the bronchi, and the results
shown in figure 9B are effectively removed by tests. The importance of the FP removal process is evident in this
example.

FIG. 5 (A) Gaussian low-pass filter visualization of 2D kernel (B) before applying and (C) after applying image
enhancement.

FIG. 6 (A) Laplacian of Gaussian filter visualization of 2D kernel (B) before applying and (C) after applying image
enhancement.
12 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

FIG. 7 Soft tissue segmentation: (A) CT image and (B) liver region of CT image.

FIG. 8 Soft tissue segmentation: (A) 3D visualization and (B) segmented vessel region of MR image.

FIG. 9 Reduction in the over-segmented artifacts (A) before applying and (B) after applying extraction.
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 13

11 Minimizing Software and Equipment Errors

11.1 INTRODUCTION TO SOFTWARE AND EQUIPMENT ERRORS
Medical computer-aided design (CAD) data obtained by segmentation and cloud point formation of image data,
such as CT and MRI images, depend on the resolution of the original image (see Appendix A). The errors that occur
in reproducing the information on irregular real human body parts with organic inflection points and a plurality of
sampled image data are inevitable. Assessing and reducing errors ultimately is related to medical image quality, and
therefore, a system that can be used for research and practical purposes needs to be established.8

11.2 SOFTWARE ERRORS

11.2.1 Background
Image processing that utilizes an interpolation algorithm can be applied to a section of an image composed of several
intervals; the algorithm can be applied at intervals with no information. To interpolate a specific section, the method
of application of geometric information and slope or curve characteristics depend on the number of shear and trail-
ing slides referenced. The error in interpolation without image information is the highest among the errors. When
there is no singular point within the slide interval and while referring to the front and rear information, this error
can be reduced to a negligible level. Singularities need to be examined by medical image analysis experts or doctors.
Errors mainly are due to the uncertainty of the image information that occurs during image acquisition and image
processing. As image processing is performed during thresholding for segmentation, an error occurs in this process.
The 3D data creation software for medical image data has been commercialized at an advanced level and is being
used to simulate medical operations for evaluating preoperative surgical treatment options with continuous research
and development and manufacturing design of surgical medical devices and implants. Thus, operator behavior
becomes a variable. In the segmentation process, a segmentation method is needed that can vary depending on the
type of the anatomical structure and a structure profile that can optimize the image processing algorithm.

11.2.2 Verification method using main inflection

To correct problems related to critical error or scale, verification can be performed by designating a main inflec-
tion part as a reference point in the previously generated mesh data by referring to a high-resolution image and
an X-ray image. The section profile is used as a verification reference. By applying the guidelines according to
the characteristics of the anatomical structure to the certification criteria of the software created by the medical
image enterprise, a basis to improve the quality of 3D medical care can be provided. Methodically, it is possible
to define error images by type and to evaluate quality through a dimensional trueness evaluation between the
corrected input results of erroneous images and the actual standard data. The evaluation methods are based on
volume difference, maximum error of coordinates, and average error for intended use.

11.2.3 Improving accuracy and precision

An additional class of software errors is generated by converting the image data from a source capture system to
DICOM PACs format to segmentation software to modeling software to print preparation software to the
MAM system software.9
The lack of an automated method for recording and sending resolution, precision, color, surface texture,
material and other key meta data affecting image quality from the image source through the software conversion
process degrades MAM device quality. An automated method is needed to maintain accuracy and precision as
close to the original image accuracy and precision as possible.
The following methodology can be used to optimize accuracy and precision of MAM-produced devices:
Replace use of the STL format with ISO/ASTM standard AMF format.10 Utilize the format’s extensible
XML meta data support to provide an integrated, accurate, and reliable means of recording and transmitting all
relevant meta data from the image source system and each software application used to create, store, segment,
edit, and output the completed 3D image file to the MAM system.
14 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

Use segmentation and modeling software that can output the file and all related resolution, accuracy, preci-
sion, and functional parameter settings saved as meta data tags in the AMF file. Use modeling software applica-
tions to assign colors, materials, surface textures, multipart assembly orientation, software parameter settings,
application version history, and any additional meta data created by each software application that may improve
the quality of the medical device.
The MAM system technician can utilize this meta data to optimize print preparation software parameter
settings, select the most appropriate MAM system for producing the medical device, and configure the MAM
system parameters to align with the meta data integrated within the image file. The MAM system service pro-
vider can use this file meta data as input to its quality management system required by the U.S. Food and Drug
Administration and the European Union for verifying and validating accuracy, precision, and reliability of
MAM-produced devices.

11.3 EQUIPMENT ERRORS

11.3.1 Background
The software error correction rate cannot reduce the error rate in an AM process. Even if the same equipment has
different characteristics, it is necessary to have guidelines on individual calibration methods for the equipment. The
calibration is performed by the phantom for which the level of X-ray absorption is similar, and which is needed to
calibrate the evaluation of each phantom model (metal phantom and simplified measurement drawing).

11.3.2 Standard computational mesh model data creation for an evaluation method
The printing result that is obtained through a series of processes is used as the phantom model entity from the
medical image data (CT and MRI) generated in the first input step. After obtaining the final output of the 3D
data creation process and AM process and scanning this output and comparing it with the phantom mesh data,
the final matching rate can be confirmed.11 In addition, the phantom model can confirm the accuracy for data
generation of medical imaging devices and optical scanning apparatus. Therefore, calibration of equipment
based on the phantom model is also possible.

11.4 TOLERANCE ERROR SITUATIONS

This section describes the tolerance for errors that may occur during the acquisition of medical image data and
the allowable range from the AM device. In actual applications, dimensional tolerance is important with regard
to the material and postprocessing:12,13,14
• data creation (segmentation),
• AM processing (mechanical tolerance),
• deformation and shrinkage (material properties), and
• post-processing (polishing tolerance).

Appendix A (Informative): Medical CAD for Additive

Manufacturing Tolerance
A.1 Purpose
The measurements of medical additive manufacturing (AM) output can provide a database of reliable dimen-
sional errors for each process. To perform this measurement, the database compares and analyzes a three-
dimensional (3D) model, AM model (scan data), and real model (scan data). Based on the database, a standard
phantom model for medical 3D printing not only can provide opportunities for service providers but also
increases credibility (see fig. A.1).
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 15

FIG. A1 Standardization of medical CAD for AM.

A.2 Procedure
A CT scan was performed to acquire DICOM data of real bones from an anatomy class. By using high-precision
scanning with the slice of 0.01 mm, a standard model data of real bones to be compared was obtained. We printed
medical 3D model data from two types of data: the CT scan and real bone scan, as shown in figure A.2.
The phantom model, which can be a standard model, can be used with real human bones. Two types of
models were prepared for the phantom model from the actual human bones. One type of model was converted
to STL by high-precision scanning of bone, and the other type of model was converted from the bone’s CT into
STL by software.
Errors occurred between the real bone and the data generated by the software during the transformation
process, and deformations due to shrinkage occurred in the material and equipment.12

FIG. A2 File comparison analysis process of three sets of data: (A) 3D model, (B) AM model scan data, and (C) real
model scan data.
16 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

The results are shown in figures A.3 to A.8, which indicate that the tolerance of color contour occurred by
merging two sets of CT data and standard model data. (The standard model achieved the highest level of high-
precision scanning possibility.) Green indicates an exact match with both sets of data, the part that is bigger
than the real bone will turn red, and the part that is smaller will turn blue.

FIG. A3 Scapula: real versus CT model deviation distribution.

ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 17

FIG. A4 Scapula: Real versus AM model deviation distribution.

18 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

FIG. A5 Scapula: CT versus AM model deviation distribution.

ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 19

FIG. A6 Humerus: Real versus CT model deviation distribution.

FIG. A7 Humerus: CT versus AM model 1 deviation distribution.

20 TECHNICAL REPORT: ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA

FIG. A8 Humerus: CT versus AM model 2 deviation distribution.

A.3 Results
These results are atypical because of the difference between the instrument resolution and the CT image quality.
The CT data on the scapula-size scale generally range from 7% to 8% of the real bone, which can reach
64.0 mm in the maximum error area. The maximum error of AM is about 60.5 mm, and the error obtained by
CT was much larger than the tolerance of the printer. See figures A.3 to A.8.

References
1. Implants for Surgery—Orthopaedic Joint Prosthesis—Part 1: Procedure for Producing Parametric 3D Bone
Models from CT Data of the Knee, ISO 19233 1 (Geneva, Switzerland: International Standards Organiza-
tion, 2017).
2. IEEE Recommended Practice for Three-Dimensional (3D) Medical Modeling, IEEE-SA P3333.2.1-2015
(Piscataway, NJ: Institute of Electrical and Electronics Engineers Standards Association, 2015).
3. D. Mitsouras, P. Liacouras, A. Imanzadeh, A. A. Giannopoulos, T. Cai, K. K. Kumamaru, E. George, et al.,
“Medical 3D Printing for the Radiologist,” RadioGraphics 35, no. 7 (2015): 1965–1988.
4. F. J. Rybicki, RSNA 2014 3D Printing (Hands-On) Training Guide (Boston, MA: Applied Image Science
Lab, 2014).
5. Additive Manufacturing—General Principles—Standard Practice for Part Positioning, Coordinates and
Orientation, ISO/ASTM 52921 (Geneva, Switzerland: International Standards Organization, 2019).
ADDITIVE MANUFACTURING FOR MEDICAL DATA: OPTIMIZED MEDICAL IMAGE DATA 21

6. H. W. Goo, S. J. Park, and S.-J. Yoo, “Advanced Medical Use of Three-Dimensional Imaging in Congenital
Heart Disease: Augmented Reality, Mixed Reality, Virtual Reality, and Three-Dimensional Printing,”
Korean Journal of Radiology 21, no. 2 (2020): 133–145.
7. D. Piretzidis and M. G. Sideris, “Stable Recurrent Calculation of Isotropic Gaussian Filter Coefficients,”
Computers and Geosciences 133 (2019): 104303.
8. M. Odeh, D. Levin, J. Inziello, F. L. Fenoglietto, M. Mathur, J. Hermsen, J. Stubbs, and B. Ripley, “Methods
for Verification of 3D Printed Anatomic Model Accuracy Using Cardiac Models as an Example,” 3D Printing
in Medicine 5, no. 6 (2019), https://doi.org/10.1186/s41205-019-0043-1
9. Health Informatics—Digital Imaging and Communication in Medicine (DICOM) Including Workflow and
Data Management, ISO 12052 (Geneva, Switzerland: International Standards Organization, 2017).
10. Specification for Additive Manufacturing File Format (AMF) Version 1.2, ISO/ASTM 52915 (Geneva,
Switzerland: International Standards Organization, 2020).
11. J.-W. Choi, J.-J. Ahn, K. Son, and J.-B. Huh, “Three-Dimensional Evaluation on Accuracy of Conventional
and Milled Gypsum Models and 3D Printed Photopolymer Models,” MDPI Materials 12, no. 21 (2019):
3499, https://doi.org/10.3390/ma12213499
12. Additive Manufacturing—Test Artifacts—Geometric Capability Assessment of Additive Manufacturing
Systems, ISO/ASTM 52902 (Geneva, Switzerland: International Standards Organization, 2019).
13. T. Lieneke, V. Denzer, G. A. O. Adam, and D. Zimmer, “Dimensional Tolerances for Additive Manufacturing:
Experimental Investigation for Fused Deposition Modeling,” Procedia CIRP 43 (2016): 286–291.
14. Z. Zhua, N. Anwera, and L. Mathieua, “Deviation Modeling and Shape Transformation in Design for
Additive Manufacturing,” Procedia CIRP 60 (2017): 211–216.
ASTM INTERNATIONAL
Helping our world work better

www.astm.org
ISBN-EB: 978-0-8031-7154-1

eBook Stock #: ISO/ASTMTR52916-EB