ISO/IEC 16466:2025

International
Standard
ISO/IEC 16466
First edition
Information technology —
2025-08
3D printing and scanning —
Assessment methods of 3D scanned
data use in 3D printing
Reference number
© ISO/IEC 2025
All rights reserved. Unless otherwise specified, or required in the context of its implementation, no part of this publication may
be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting on
the internet or an intranet, without prior written permission. Permission can be requested from either ISO at the address below
or ISO’s member body in the country of the requester.
ISO copyright office
CP 401 • Ch. de Blandonnet 8
CH-1214 Vernier, Geneva
Phone: +41 22 749 01 11
Email: copyright@iso.org
Website: www.iso.org
Published in Switzerland
© ISO/IEC 2025 – All rights reserved
ii
Contents Page
Foreword .iv
Introduction .v
1 Scope . 1
2 Normative references . 1
3 Terms, definitions and abbreviated terms . 1
3.1 Terms and definitions .1
3.2 Abbreviated terms .2
4 Assessment . . 3
4.1 General .3
4.1.1 Background .3
4.1.2 Published standards and metrics of assessment.4
4.2 Types of errors in image segmentation of 3D scanned data .10
4.2.1 General .10
4.2.2 Typical examples of each type of error .10
5 Approach to assessments . 10
5.1 General .10
5.2 Region of Intertest/Volume of Interest .11
5.3 Image enhancement and image normalization .11
5.4 Surface modelling and pre-processing (smoothing and averaging) .11
6 Assessment for segmentation of 2D images.11
6.1 General .11
6.2 Workflow and product quality .11
6.3 Assessment methods for image-based modelling/segmentation phase . . 12
6.3.1 Region-based measure/spatial overlap based metrics; Sensitivity, Specificity,
False-positive rate, False negative rate FNR, F-Measure FMS, Dice similarity
coefficient, Jaccard index . 12
6.3.2 Volume-based measure; Volume similarity, Volume overlap error, Volume
difference . 13
6.3.3 Probabilistic Distances Between Segmentation/Cross-correlation matrix
measure; Interclass correlation (ICC), AUC, Cohen kappa coefficient . 13
6.3.4 Distance-based measure/Spatial distance-based metrics; Hausdorff distance
(HD, Maximum Surface Distance), Mahalanobis distance . 13
7 Assessment for 3D modelled images .13
7.1 General . 13
7.2 Surface Distance-based Measure: Hausdorff distance, Average distance, Mean Absolute
Surface Distance, Mahalanobis distance .14
8 Choosing the most suitable metric for assessing image segmentation . 14
Annex A (informative) Preparation of Dataset for Assessment for image based modelling .15
Annex B (informative) Assessment for image based modelling of Craniofacial 3D images .16
Annex C (informative) Assessment for Orbital Segmentation . 19
Annex D (informative) Tools to evaluate the quality of image segmentation .22
Bibliography .24

© ISO/IEC 2025 – All rights reserved
iii
Foreword
ISO (the International Organization for Standardization) and IEC (the International Electrotechnical
Commission) form the specialized system for worldwide standardization. National bodies that are
members of ISO or IEC participate in the development of International Standards through technical
committees established by the respective organization to deal with particular fields of technical activity.
ISO and IEC technical committees collaborate in fields of mutual interest. Other international organizations,
governmental and non-governmental, in liaison with ISO and IEC, also take part in the work.
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 document 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 or www.iec.ch/members_experts/refdocs).
ISO and IEC draw attention to the possibility that the implementation of this document may involve the
use of (a) patent(s). ISO and IEC take no position concerning the evidence, validity or applicability of any
claimed patent rights in respect thereof. As of the date of publication of this document, ISO and IEC had not
received notice of (a) patent(s) which may be required to implement this document. However, implementers
are cautioned that this may not represent the latest information, which may be obtained from the patent
database available at www.iso.org/patents and https://patents.iec.ch. ISO and IEC shall not be held
responsible for identifying any or all such patent rights.
Any trade name used in this document is information given for the convenience of users and does not
constitute 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
Organization (WTO) principles in the Technical Barriers to Trade (TBT) see www.iso.org/iso/foreword.html.
In the IEC, see www.iec.ch/understanding-standards.
This document was prepared by Joint Technical Committee ISO/IEC JTC 1, Information technology.
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 and
www.iec.ch/national-committees.

© ISO/IEC 2025 – All rights reserved
iv
Introduction
This document was developed in response to the need for quality management of 3D printing and scanning
technology through the use of information and communication technology (ICT).
3D scanning is the process of scanning a real-world object or environment to collect data on its shape and
possibly its style attributes. The main purpose of 3D scanning is for generating high-precision digital 3D models.
A 3D scanner can be based on many different technologies, each with its own purposes and targets,
limitations, and advantages. There could be many limitations in each type of target object that will be
digitized. For example, optical technology often encounters many difficulties with dark, shiny, reflective, or
transparent objects. Another example, as for computed tomography scanning, structured-light 3D scanners,
and LiDAR technology, the generation of digital 3D models requires the use of non-destructive internal
scanning technology.
Despite the rapid growth of 3D scanning applications, the accuracy, precision and reproducibility of
generated 3D models from 3D scanned data have not been thoroughly investigated. Especially if 3D scanned
data are used for 3D printing, their accuracy and precision are critical. Inaccuracies can arise due to errors
that occur during the imaging, segmentation, postprocessing, and 3D printing steps. The total accuracy,
precision, and reproducibility of 3D printed models are affected by the sum of errors introduced in each step
involved in the creation of the 3D models.
For the spreading of 3D printing applications, it is necessary to review and evaluate the various factors in
each step of the 3D model printing process that contribute to 3D model inaccuracy, including the intrinsic
limitations of each printing technology.
In this context, it is important to evaluate the overall process of data processing. In order to minimize
cumulative errors throughout the 3D printing life cycle using 3D scanned data, it is important to evaluate
and correct initial errors. By identifying and addressing these initial inaccuracies, the impact of errors
occurring during the 3D printing process can be greatly reduced. In addition, the method used to evaluate
3D scan data for 3D printing is also essential.
There are many algorithms for 3D scanned data such as semi-automatic segmentation, deformable model-
based segmentation, and Convolutional Neural Network based segmentation. There are several well-known
errors during image-based modelling of Region of Interest (ROI), which are over segmentation, under
segmentation, outlier, inaccurate contour, and malalignment. Even though there are more than twenty
metrics for evaluating 3D image segmentation, there is no consistent definition of metrics and suitable
combination of assessment metrics for 3D printing.
Segmentation assessment is the task of comparing two segmentations by measuring the distance or
similarity between them, where one is the segmentation to be assessed and the other is the corresponding
ground truth segmentation.
There are three major requirements (accuracy, precision, and efficiency) of assessment for 3D scanned
data. Accuracy is the degree to which the segmentation results agree with the ground truth segmentation.
Precision is a measure of repeatability. Efficiency is mostly related with time.
This document proposes assessment methods for 3D scanned data to evaluate and enhance the quality of 3D
printing models while minimizing errors.

© ISO/IEC 2025 – All rights reserved
v
International Standard ISO/IEC 16466:2025(en)
Information technology — 3D printing and scanning —
Assessment methods of 3D scanned data use in 3D printing
1 Scope
This document specifies methods and metrics for assessing the accuracy and precision of 3D scanned data
for use in 3D printing, throughout the full 3D printing lifecycle.
This document focuses mainly on 3D scanned data from computed tomography. Computed tomography can
acquire information concerning the internal structures, regional density, orientation and/or alignment of
scanning objects, as well as their shape and appearance.
This document is applicable to the assessment of image-based modelling, segmentation, and 3D models.
This document is not intended to assess the 3D printed product itself.
2 Normative references
There are no normative references in this document.
3 Terms, definitions and abbreviated terms
For the purposes of this document, the following terms and definitions apply.
ISO and IEC maintain terminological databases for use in standardization at the following addresses:
— ISO Online browsing platform: available at https:// www .iso .org/ obp
— IEC Electropedia: available at https:// www .electropedia .org/
3.1 Terms and definitions
3.1.1
assessment
act of judging or deciding the amount, value, quality, or importance of something, or the judgment or decision
that is made
3.1.2
contour
shape of an anatomical structure or part of body, especially its surface or the shape formed by its outer edge
3.1.3
enhancement
process of improving the quality, amount, or strength of something
3.1.4
Euclidean distance
length of a line segment between the two points in Euclidean space
3.1.5
normalization
changing of the values of numeric columns in the dataset to use a common scale, without distorting
differences in the ranges of values

© ISO/IEC 2025 – All rights reserved
3.1.6
reference image
image that is assumed to have perfect quality
3.1.7
region of interest
ROI
boundary of a specific object defined in the image
[SOURCE: ISO/IEC 3532-2:2024, 3.7]
3.1.8
segmentation
process of separating the objects of interest from their surroundings
Note 1 to entry: segmentation can be applicable to 2D, 3D, raster or vector data.
Note 2 to entry: segmentation method: terms and definition standardized by ISO [ISO 13322-1:2014(en)].
3.1.9
segmented image
separated image extracted from original image with ROI
3.1.10
ground truth
ground truth label
correct answer of the training set for segmentation based on supervised learning
[SOURCE: ISO/IEC 3532-2:2024, 3.6]
Note 1 to entry: “The term for 3.6 of ISO/IEC 3532-2:2024 is “ground-truth label”, the “ground truth” is added in this
document.
3.2 Abbreviated terms
ARI adjusted rand index
AUC area under ROC curve
AVD average Hausdorff distance
CT computed tomography
DICOM digital imaging and communications in medicine
DSC Dice similarity coefficient
FMS F-measure
FN false negative
FNR false negative rate
FP false positive
FPR false positive rate
GCE global consistency error
HD Hausdorff distance
ICC interclass correlation
© ISO/IEC 2025 – All rights reserved
JAC Jaccard index
KAP Cohens kappa measure
MHD Mahalanobis distance
MI mutual information
PBD probabilistic distance
RI Rand index
3D three dimensional
TN true negative
TNR true negative rate
TP true positive
TPR true positive rate
2D two dimensional
VOI variation of information
VS volumetric similarity
4 Assessment
4.1 General
4.1.1 Background
Precision, accuracy and efficiency of image processing of 3D scanned data are essential for the application of
3D printing. At the very beginning of image processing, the segmentation methods should be high-precision
and accurate. Therefore, assessing the accuracy and quality of the segmentation process and related
algorithms is very important.
Image segmentation has different quality aspects.
a) generality for evaluation; the assessment method should be applicable to most of the assessment of
image segmentation for 3D scanned data processing.
b) subjective versus objective; currently, mostly subjective assessments exist, conducted by experienced
image experts or image processing experts for 3D printing. Even though many objective assessment
methods also exist, those are not standardized for 3D scanned data for the 3D printing model.
c) complexity for evaluation; most of the 2D images from computed tomography are pixel or voxel based
planar images. In the context of the 3D printing model, image processing reconstructs these images
into 3D images which are mostly STL file format or surface mesh format. Most assessment methods
for image segmentation are based on overlap, volume, information theory, probabilistic distance,
and pair counting. There are more than twenty evaluation metrics for these assessments for image
segmentation. In general, practice in automatic segmentation algorithms and manual segmentation
are mixed. Additionally, for 3D printing, volume and alignment of 3D scanned data are important. So,
current evaluation metrics cannot afford to properly assess the image segmentation of 3D scanned data
for 3D printing.
d) evaluation requirements for reference images; current image segmentation algorithms use ‘feature-
based or texture-based’ methods. Those algorithms are assessed using 'reference images,' known as

© ISO/IEC 2025 – All rights reserved
'Gold standards' or 'Ground truth.' Reference images are confirmed by highly experienced image
experts.
Annex A provides a preparation of dataset for assessment for image based modelling as well as technical
issues of CT scanning.
4.1.2 Published standards and metrics of assessment
4.1.2.1 ISO/IEC 3532 series
The ISO/IEC 3532 series provide general requirements and technical guidelines for medical image-based
modeling on medical 3D printing.
— ISO/IEC 3532-1:2023 provides general requirements for the use of image-based modeling for 3D printing
particularly in medicine. By following these requirements, professionals and manufacturers can ensure
that their 3D printed parts and devices and 3D models are safe, effective and reliable.
— ISO/IEC 3532-2:2024 provides guidelines and requirements for the segmentation of medical image data
in the context of 3D printing and scanning in addition to important guidance on the selection and use of
segmentation algorithms and techniques. The key perspective of ISO/IEC 3532-2:2024 is the importance
of considering patient-specific factors in segmentation, such as anatomy, pathology, and medical history.
This includes individual variations in shape, size, and structure, as well as any anomalies or abnormalities
that may affect segmentation accuracy.
4.1.2.2 Similarity metrics for assessments
Similarity metrics for assessment focus on the methods used to measure similarity between different
datasets or models. Comparing these metrics evaluates the degree of agreement or discrepancy between a
reference model and the model being assessed, ensuring high-quality and reliable outcomes.
a) Dice Similarity Coefficient (DSC) is the most widely used metric when validating image segmentation,
and it is defined as:
2n
TP
S =
2nn++n
TP FP FN
where
S is calculated Dice Similarity Coefficient (DSC);
n is representing number of occurrences.
b) Jaccard Coefficient (JAC) is defined as the intersection between them divided by their union, and it is
defined as:
n
TP
J =
nn++n
TP FP FN
where
J is calculated Jaccard Coefficient (JAC);
n is representing number of occurrences.
c) Area under ROC Curve (one system state) (AUC) is defined as: The ROC curve (Receiver Operating
Characteristic) is the plot of the true positive rate against the false positive rate. It normally assumes
more than one measurement. AUC is defined by the measurement point and the lines TPR = 0 and FPR =
1, which is given by:
1  n n 
FP FN
A =−1 +
 
2 nn+ nn+
 
FP TN FN TP
© ISO/IEC 2025 – All rights reserved
where
A is calculated Area under ROC Curve (one system state), AUC;
n is representing number of occurrences.
d) Cohen Kappa Coefficient (KAP) is a measure of agreement between two samples. As an advantage
over other measures, KAP takes into account the agreement caused by chance, which makes it more
robust. KAP is given by:
PP−
ac
K =
1−P
c
where
K is calculated Cohen Kappa Coefficient, KAP;
P is the agreement between the samples;
a
P is the hypothetical probability of chance agreement.
c
e) Rand Index is a measure of similarity between clusterings. One of its important properties is that it is
not based on labels and thus can be used to evaluate clusterings as well as classifications. The RI between
two segmentations s (the ground truth segmentation) and s (the test segmentation) is defined as:
g t
ab+
R =
ab++cd+
where R is calculated Rand Index, RI (s , s );
g t
aT=− [ PT()PF11+−PF()PT+−PT()NF11+−NF()N ;
22 22
 
bT=+ ()PFNT++()NFPT−+PTNF++PFN ;
()
 
22 22 22
 
cT=+()PFPT++()NFNT−+PTNF++PFN ;
()
 
nn()−1
d= −+()ab+c .
f) Adjusted Rand Index (ARI) is a modification of the Rand Index that considers a correction for chance.
The ARI can be expressed by the four cardinalities as:
2()ad−bc
I =
cb++ 2ad++ad cb+
()()
where I is calculated Adjusted Rand Index, ARI.
Note 1 to entry The symbols “a” - “d” within this formula are defined in the same way here as in the previous
formula, Rand Index, RI.
g) Interclass Correlation (ICC) is a measure of correlations between pairs of observations that don’t
necessarily have an order or are not obviously labeled. It is common to use the ICC as a measure of
conformity among observers. It usually uses as a measure of consistency between two segmentations.
ICC is given by:
σ
s
C=
σσ+
s 
© ISO/IEC 2025 – All rights reserved
where
C is calculated Interclass Correlation, ICC.
σ is variance caused by differences between the segmentations;
s
σ is variance caused by differences between the points in the segmentations.

h) Volumetric Similarity Coefficient (VS) is a measure that considers the volumes of the segments to
indicate similarity and the absolute volume difference divided by the sum of the compared volumes. The
Volumetric Similarity (VS) is defined as 1 − VD where VD is the volumetric distance. That is
vv−
FN FP
V =−1
2vv++v
TP FP FN
where
V is calculated Volumetric Similarity Coefficient, VS;
v is representing calculated volume of interest.
i) Mutual Information (MI) between two variables is a measure of the amount of information one
variable has about the other. Knowing the value of another variable helps reduce the uncertainty of a
variable. The MI as a similarity measure between image segmentations in particular, are calculated
based on regions (segments) instead of individual pixels.
4.1.2.3 Distance metrics for assessments
Distance metrics for evaluation are mathematical metrics used to quantify similarities or differences
between two data points or models. These metrics are critical for assessing the accuracy and reliability of a
3D model by comparing it to baseline 3D scanned data.
a) Hausdorff Distance (optionally in voxel or millimeter) (HD) between two finite point sets A and B is
defined by:
Hh=max,AB ,, hB A
()() ()
where
H is calculated Hausdorff Distance (optionally in voxel or millimeter), HD (A, B);
h(A, B) is called the directed Hausdorff distance and given by:
hA(),mBa=−axmin b
aA∈ bB∈
where ab− is a given norm, e.g. Euclidean distance.
The HD is generally sensitive to outliers. Because noise and outliers are common in medical
segmentations, it is not recommended to use the HD directly.
b) Average Hausdorff Distance (optionally in voxel or millimeter) (AVD) is known to be stable and less
sensitive to outliers than the HD. It is defined by:
Dd=max,()()AB ,, dB()A
Where
D is calculated Average Hausdorff Distance (optionally in voxel or millimeter), AHD (A, B);
d(A, B) is the directed Average Hausdorff distance that is given by:
dA,mB =−in ab
()
∑
N bB∈
aA∈
c) Balanced Average Hausdorff Distance is given by:

© ISO/IEC 2025 – All rights reserved
D D
 
gs,, s g
D =+ /2
 
b
 
n n
g g
 
where
D is calculated Balanced Average Hausdorff Distance, Balanced AHD;
b
D is the directed average Hausdorff distance from ground truth to segmentation;
g,s
D is the directed average Hausdorff distance from segmentation to ground truth;
s,g
n is the number of voxels in the ground truth;
g
n is the number of voxels in the segmentation.
s
d) Mahalanobis Distance (MHD) between two points in a point cloud, in contrast to the Euclidean
distance, takes into account the correlation of all points in the point cloud containing the two points.
The MHD between the points x and y in the point cloud A is given by
T
−1
Mx=−()yP ()xy−
s
where
M is calculated Mahalanobis Distance, MHD (x, y);
-1
P is the inverse of the covariance matrix S of the point cloud;
s
T is the matrix transpose.
NOTE x and y are two points in the same point cloud, but in the validation of image segmentation, two point
clouds are compared.
e) Variation of Information (VOI) measures the amount of information lost (or gained) when changing
from one variable to the other. VOI measure for comparing clusterings partitions.
f) Global Consistency Error (GCE) is an error measure between two segmentations. Let Rs(s, x) be
defined as the set of all voxels that reside in the same region of segmentation s where the voxel x resides.
For the two segmentations s1 and s2, the error at voxel x, E (s1, s2, x) is defined as:
Rs(),,xR{ sx
()
st sg
E =
Rs ,x
()
st
where
R is the set of all voxels that reside in the same region of segmentation.
s
E is calculated Error, E (s , s , x).
t g
Note that E is not symmetric.
The global consistency error (GCE) is defined as the error averaged over all voxels and is given by:
n n
 
1  
E = minE ss,, xE,,ss ,,x
() ()
 
gt∑∑gi gt i
n
 
 i i 
where E is calculated global consistency error, GCE (s , s ).
g t g
g) Coefficient of Variation (CV) is a measure of the ratio of the standard deviation to the mean. The metric
does not measure the correctness of segmentation, only the variability of segmentation algorithm on
different datasets.
σ
V =
μ
© ISO/IEC 2025 – All rights reserved
where
V is calculated Coefficient of Variation, CV;
σ is the standard deviation;
μ is the mean.
h) Probabilistic Distance (PBD) is a measure of distance between fuzzy segmentations. Given two fuzzy
segmentations, A and B, then the PBD is defined by:
∫−pp
AB
p =
I
2∫ p
AB
where
p is calculated Probabilistic Distance, PBD (A, B);
p (x) and p (x) are the probability distributions representing the segmentations;
A B
p is their pooled joint probability distribution.
(A, B)
Applied on s and s , the PBD is defined as
g t
fx − fx
() ()
∑ gt
x
p =
2 fx fx
() ()
∑ gt
x
where
p is calculated Probabilistic Distance, PBD (s , s );
2 g t
x is representing the specific points that make up 3D scanned data;
fx is the assignment function providing the membership of the object x in the subset s ;
()
g g
fx is the assignment function provides the membership of the object x in the subset s
()
t t,
4.1.2.4 Classic measures
a) Sensitivity (Recall, true positive rate, TPR):
n
TP
T =
nn+
TP FN
where
T is calculated true positive rate, TPR;
n is representing number of occurrences;
b) Specificity (true negative rate, TNR):
n
TN
N =
nn+
TN FP
where
N is calculated true negative rate, TNR;
n is representing number of occurrences;
c) Precision is called the positive predictive value (PPV):
n
TP
V =
PP
nn+
TP FP
© ISO/IEC 2025 – All rights reserved
where
V is calculated the positive predictive value, PPV;
PP
n is representing number of occurrences;
d) F-Measure is a trade-off between PPV (precision) and TPR (recall).
2⋅⋅VR
PP TP
F =
VR+
PP TP
where
F is calculated F-Measure, FMS;
V is calculated the positive predictive value, PPV;
PP
R is calculated true positive rate, TPR;
TP
FMS is mathematically equivalent to DSC.
e) Accuracy is used to describe the closeness of a measurement to the true value. When the term is applied
to sets of measurements of the same measurement, it involves a component of random error and a
component of systematic error. In this case trueness is the closeness of the mean of a set of measurement
results to the actual (true) value and precision is the closeness of agreement among a set of results.
f) Fallout (false positive rate, FPR)
n
FP
Q =
nn+
FP TP
where
Q is calculated false positive rate, FPR;
n is representing number of occurrences.
g) True positives (in voxel), true negatives (in voxel), false positives (in voxel), and false negatives
(in voxel) shall be defined as: For two crisp partitions (segmentations) sg and st, the confusion matrix
consists of the four common cardinalities that reflect the overlap between the two partitions, namely TP,
FP, FN, and TN. These cardinalities provide for each pair of subsets i∈sg and j∈st the sum of agreement
m between them. That is:
ij
X
i j
mf= ()xf ()x
ij ∑ g r t r
r=1
where
TP = m , FP = m , FN = m , and TN = m
11 10 01 00;
x is representing the specific points that make up 3D scanned data;
fx() is the assignment function provides the membership of the object x in the subset s ;
g g
fx() is the assignment function provides the membership of the object x in the subset s
t t,
© ISO/IEC 2025 – All rights reserved
4.2 Types of errors in image segmentation of 3D scanned data
4.2.1 General
There are four types of image segmentation errors, namely the quantity (number of segmented objects), the
area of the segmented objects, the contour (degree of boundary match), and the content (existence of inside
holes and boundary holes in the segmented region). Assessment for image segmentation is the comparison
of two segmentations by measuring the distance or similarities between them (to be evaluated vs reference
segmentation). The requirements of image segmentation evaluation include the accuracy in measuring the
degree to which segmentation results match the ground truth segmentation, the precision in evaluating the
repeatability of the results, and the efficiency primarily associated with the time required for segmentation.
For 3D printing using 3D scanned data over segmentation, under segmentation, outlier (additional small
segmented objects outside the main object), contour (the delineation of the boundary), and alignment (the
general position of the segmented object) should be considered and assessed.
4.2.2 Typical examples of each type of error
— Over-segmentation occurs when an object separated from the background is segmented larger than
the object itself and further segmented into surrounding similar regions.
— Under-segmentation occurs when an object separated from the background is segmented smaller than
the object itself.
— Outliers are relatively small wrongly segmented regions outside (normally far from) the segment.
Metrics sensitive to outliers over-penalize them. When outliers are not harmful, metrics with sensitivity
to outliers, such as the HD, should be avoided.
— Contouring errors or boundary errors depend on the goal of the segmentation, i.e. whether the exact
delimitation of the boundary is important or not. Anatomy structures that are segmented can be of
different grades of complexity in terms of boundary delimitation. They can vary from simple and smooth
shapes, like a kidney, to irregular shapes, like tumors, but also branched and complex like the vessels
of the eye retina. Image processing of 3D scanned data for 3D printing requires that the contours or
boundaries remain accurate and consistent. Many of file formats for 3D printing, such as STL, AMF, 3MF,
are based on surface mesh.
— Alignment and the extent are more important than the boundary for estimating the location and the
size or general shape of a structure. Especially for 3D printing alignment is very important and critical
property for reliable data processing from 3D scanned data for 3D printing.
5 Approach to assessments
5.1 General
User needs for quality include requirements for image-based modelling in use in specific contexts of use.
These identified needs can be considered when specifying assessment methods using modelled image
quality characteristics and sub characteristics.
The quality for image-based modelling especially of segmentation can be assessed by measuring 2D
properties (typically ‘overlap area-based metric’ and ‘distance-based metric’), or by measuring 3D properties
(typically surface distance, shape, and alignment), or by measuring quality in use properties (when the
modelled image is in 3D printing or simulated use). Appropriate 2D properties of the modelled images are a
prerequisite for achieving the required 3D properties or requirements for achieving quality in use.
It is recommended, where possible, to use 2D properties that have a strong relationship with the target 3D
properties so that they can be used to predict the values of 3D properties.
Some assessments shall be normalized against the target value specified in a requirement specification,
a design specification, or a user documentation. Such target value should be determined and required as
the threshold by developers or maintainers to improve architecture, design, implementation, assembles,

© ISO/IEC 2025 – All rights reserved
operational procedures, user interface or performance of the image-based modelling. The target value
is also able to be specified as one of agreed requirements by acquirers and suppliers to specify quality
requirements or to examine conformance for acquisition. A requirements specification should not be
changed and revised during image based modelling in such a way as to affect the assessment methods based
on it. Accordingly, operators of image-based modelling are expected to continuously evolve and revise the
requirements specification and to apply assessment methods not at once but iteratively during image-based
modelling.
Selecting the most suitable metrics to assess image segmentation depends on the data being segmented and
the goal of the segmentation task.
Two facts should be considered to select an assessment method:
— The first is that effectiveness metrics can be biased towards or against properties of the images being
segmented, meaning that particular metrics over-penalized or over-reward segmentations given
particular properties.
— The second fact is that selecting the best evaluation metrics can be subject to the segmentation goal
which means that the bias towards/against a particular property of the data can be differently important
depending on the segmentation goal.
The selected assessment method should meet the context dependency by allowing individual weighting
of the influence of each property according to its importance where this is known, which increases the
effectiveness of the method.
5.2 Region of Intertest/Volume of Interest
A small ROI should be considered when the smallest dimension of the segment, i.e. min (length, width,
height), is significantly less than the corresponding dimension of the grid on which the image is defined.
For small segments, the comparable alignment error shall be expected in magnitude with the segment size,
which results in the probability of small (or zero) overlap being high.
5.3 Image enhancement and image normalization
In a typical situation of image based-modelling image enhancement and normalization for efficient and
consistent image segmentation shall be performed prior to main ROI extraction. This image processing will
affect the original image properties and the expected results of image-based modelling.
5.4 Surface modelling and pre-processing (smoothing and averaging)
After extracting ROI, the segmented image typically undergoes smoothing or averaging for further image
based modelling. This image processing will affect the surface of the extracted image and local and global
geometry of extracted objects.
6 Assessment for segmentation of 2D images
6.1 General
Metric sensitivities are challenges in defining metrics. Sensitivity to particular properties could prevent
the discovery of particular errors or lead to over- or under-estimating them. A standard assessment tool
for image segmentation should be defined which standardizes not only the metrics to be used, but also the
definition of each metric.
6.2 Workflow and product quality
ISO/IEC 3532-1:2023 describes the generation of the 3D model and printing from 3D scanned data workflow
and its requirements. Figure 1 shows how the major phases defined in ISO/IEC 3532-1:2023 can be classified

© ISO/IEC 2025 – All rights reserved
into three task units. It also shows the 3D printing quality lifecycle and how quality measures can be related
to the overall workflow.
Figure 1 — Key tasks of 3D printing workflow [SOURCE: ISO/IEC 3532-1:2023, 4.1.1]
It might be assumed that the total error/accuracy/precision of a 3D printed product approximates the
sum of the error/accuracy/precision measured from each key tasks. Therefore, for each unit of work,
errors, accuracy, and precision might be evaluated and improved to improve overall quality. [SOURCE:
ISO/IEC 3532-2:2024]
The modeling task includes the three phase to create a 3D model for 3D printing from 3D scanned data.
Modeling task's goal is to make the best precisible 3D model for patient.
— Image acquisition phase: In the image acquisition phase, image data are acquired from imaging devices
such as CT.
— Segmentation phase: In the segmentation phase, the acquired images are segmented to fit the design
purpose. ISO/IEC 3532-2:2024 is for this segmentation phase.
— 3D modelling phase: In the 3D modelling phase, the segmented ROI is made into a 3D model optimized
for 3D printing.
The goal of quality management activity in the modeling task shall be to measure and enhance the model
quality.
6.3 Assessment methods for image-based modelling/segmentation phase
6.3.1 Region-based measure/spatial overlap based metrics; Sensitivity, Specificity, False-positive
rate, False negative rate FNR, F-Measure FMS, Dice similarity coefficient, Jaccard index
There are very peculiar features of the thin and thick bony structure of the orbital wall of human. Very
thin medial wall of orbit has more background pixels than foreground pixels. This is a significantly smaller
size for segment than background, so that it is comparable in magnitude with the expectation of the
alignment error, therefore, all metrics based on the overlap based metrics, as well as volume based metrics
are not suitable. Small segments are those with at least one dimension being significantly smaller than the
corresponding dimension of the grid on which the image is defined (e.g. less than 5 % of the corresponding
grid dimension). So specific assessment metrics for small structure and large structure (especially in solid
organs, liver, kidney, and heart) should be recommended.
There is an inverse relation between segment size (relative to the grid size) and the expectation value of the
alignment error, which directly follows from the degree of freedom for the segment location being higher
when the segment is small.
Segments that are small in only one dimension (planar shape) or small in two dimensions (linear shape)
can cause the same effect (i.e. the expectation value of the alignment error is comparable with the smallest
dimension). To illustrate this effect, consider comparing two lines using DSC (weighted Dice Similarity
Coefficient). Assume that the lines have almost exact match, but the overlap is zero. Here, the DSC provides
the same value (zero) for these two lines and for another two lines that are far from each other. The same
holds for two planes or two points.

© ISO/IEC 2025 – All rights reserved
6.3.2 Volume-based measure; Volume similarity, Volume overlap error, Volume difference
Volume similarity only compares the volume of the segments in the automatic segmentation with the volume
in the ground truth, which implicitly assumes that the segments are optimally aligned.
6.3.3 Probabilistic Distances Between Segmentation/Cross-correlation matrix measure; Interclass
correlation (ICC), AUC, Cohen kappa coefficient
These are to measure the comparison of reference segmented images and to be assessed.
6.3.4 Distance-based measure/Spatial distance-based metrics; Hausdorff distance (HD, Maximum
Surface Distance), Mahalanobis distance
For similarity assessment of image segmentation, spatial distance based methods are mostly used. These
are recommended for the segmentation overall accuracy and the boundary delineation (contour). Specific
structures that are segmented can be of different grades of complexity in terms of boundary delimitation.
They can vary from simple and smooth shapes.
Hausdorff distance is sensitive to outliers, so it is not recommended to use HD directly. The average distance,
or the Average HD, is the HD averaged over all points. The AVD is known to be stable and less sensitive to
outliers than the HD.
EXAMPLE 1
Alignment and shape A star is compared with a circle. The same star is then compared with another star that
has the same shape and dimensions, but slightly
...