Note: Descriptions are shown in the official language in which they were submitted.
~ W 0 94/14132 2 1~ 12 ~ O PCTIGB93/02504
NON-INVASIVF MEDICAL SCANNING
BACKGROUND OF THE INVENTION
Field of the Invention
This invention relates to non-invasive medical scanning.
3escri~tion of the Prior Art
Non-invasive medical scAnning, such as ultrasound scanning, is
used in the imaging of the interior of the human body. In the case of
ultrasound scanning, ultrasonic vibrations generated by a hand-held
transducer are transmitted into the body through the skin or mucous
membranes, and are reflected back to the transducer from tissues of
different densities within the body. As described in the book "Physics
and Instrumentation of Diagnostic Medical Ultrasound" (P. Fish, John
Wiley & Sons, 1990), an image of a cross-section through the body can
be built up by analysing the relative amplitudes and delays of the
reflected vibrations.
One particular use which has been made of ultrasound scAnning is
the examination of the fetus in the mother's uterus. A skilled
operator can determine the orientation, gestational age and general
condition of the fetus from its ultrasound image. Also, some fetal
abnormalities, such as cardiac or renal abnormalities, can be diagnosed
by the operator. A difficulty is that some fetal abnormalities which
would be evident from the ultrasound image may be overlooked because
they occur so rarely that an operator would lack sufficient experience
to be able readily to recognise those abnormalities.
S~MMARY OF THF INVENTION
It is an object of the invention to improve the detection of
medical abnormalities from non-invasively scanned images.
This invention provides a non-invasive medical scanning apparatus
comprising: scanning means for non-invasively generating an image of at
least an interior region of a subject to be e~Amine~; and means for
detecting a quantitative measure indicative of the shape of said image,
said quantitative measure having a correlation with the presence of a
medical abnormality in said subject.
This invention addresses the problem that some rare medical
abnormalities may be evident from a non-invasively scanned image of the
patient, but the recognition by an operator of these abnormalities may
W O 94/14132 PCT/GB93/02504
2lsi28a
be difficult (through lack of operator experience) and time consuming.
As an example, the condition of spina bifida in a fetus is
indicated by a deformation of the fetal head. This deformation (known
as the "lemon" sign) shows up on an ultrasound scan of the fetus within
the mother's uterus, but the fact that spina bifida occurs in only
about one in one thousand term births means that (a) the scanning
operator would have to, on average, examin~ lOOO ultrasound images to
detect one case of spina bifida; and (b)~many operators would only be
likely to see one or two cases of spina, ~ifida per year, and so would
not develop the experience to be ab ~ to detect the deformation or
abnormality.
The invention solves this problem by detecting a quantitative
measure indicative of the shape of the non-invasively scanned image and
correlated with the presence of a medical abnormality. This process
can be performed automatically when each image is generated, and can be
used (for example) to generate an alarm signal to prompt further
medical investigations, such as more detailed scanning or further
testing by an expert in that medical field.
The quantitative measure can be used in various ways in order to
assist an operator to be made aware of the medical abnormality in
question. For example, in one preferred embodiment, the apparatus
comprises means for displaying a numerical value indicative of said
quantitative measure. In another preferred embodiment, the apparatus
comprises means for detecting whether said quantitative measure lies in
a range indicative of possible presence of said medical abnormality.
In order to attract the operator's attention in the case of a
detection of a possible medical abnormality, it is preferred that the
apparatus comprises means for generating an alarm signal in response to
a detection that said quantitative measure lies in said range
indicative of said possible medical abnormality.
In a preferred embodiment the apparatus comprises an alphanumeric
display; and in said alarm signal comprises an alarm message for
display on said alphanumeric display.
Although the quantitative measure can be compared with static
data representing an overall population, it is preferred that data
derived from the scanned images can be archived, in order that future
images may be compared with the archived data. Accordingly, in a
~ W O 94/14132 215 12 8 0 PCT/GB93/02504
preferred embodiment, the apparatus comprises means for storing data
relating to previous occurrences of said medical abnormality; and means
for comparing said quantitative measure with said stored data relating
to said previous occurrences of said medical abnormality. In
particular, it is preferred that said stored data comprises data
indicative of a statistical distribution of said quantitative measure
in said previous occurrences of said medical abnormality.
In a preferred embodiment the apparatus comprises means for
deriving a likelihood ratio from said detected quantitative measure and
said stored data, said likelihood ratio indicating a relative
likelihood of occurrence of said medical abnormality in a current
subject. Preferably the apparatus also comprises means for displaying
said likelihood ratio for said current subject.
The likelihood ratio may simply be displayed, for evaluation by
the operator. However, it is preferred that the evaluation is
performed at least in part automatically. To this end it is preferred
that the apparatus comprises means for detecting whether said
likelihQod ratio lies in a predetermined range; and means for
generating an alarm signal if said likelihood ratio lies in said
predetermined range.
The likelihood ratio is preferably combined with the "prior risk"
of the occurrence of the medical abnormality. To this end, it is
preferred that the apparatus comprises input means for user input of a
prior risk value, said prior risk value indicating a statistical
occurrence of said medical abnormality; and means for deriving an
absolute risk value by deriving a product of said prior risk value and
said like~ih~od ratio. For example, if the prior risk is 1:1000 (1
case in 1000 population) and the likelihood ratio is 3, then the
absolute risk value would be 3:1000.
Preferably the analysis of the scanned image is combined with
analysis of other tests which can support a diagnosis of the
abnormality in question. It is thus preferred that the apparatus
comprises input means for input of test data indicative of a further
test for said medical abnormality. It is also preferred that the
apparatus comprises means for combining said test data with said
quantitative measure and said stored data, to generate said like~ihood
ratio.
wo 94/14132 ~ ~ S ~ 2 ~ ~ PCT/GB93/02504 ~
Although the further tests could take many forms, it is preferred
that said test data is indicative of a result of a biochemical test
performed on said subject.
The input means could be as simple as a data entry keyboard.
However, to allow for the electronic transfer of test data from other
apparatus, it is preferred that the input means comprises an electronic
data interface. ~ ~
The derivation of the likelihood ratio is subject to the effects
of random noise in the quantitative measure and (if applicable) the
other tests performed. These effects can be reduced by allowing for
variable factors relating to the current subject. It is therefore
preferred that the apparatus comprises user input means for user input
of subject data specifying one or more physical attributes of said
subject, said means for deriving said likelihood ratio being responsive
to said subject data.
Preferably said subject data comprises at least one of:
(i) said subject's age; and
(ii) said subject's weight.
It is preferred that the apparatus comprises means for digitally
enhancing said image. This may comprise means for high-pass spatially
filtering said image. Means for detecting portions of said image
having at least a predetermined luminance level may also be employed to
detect portions of interest in the images. In order to detect edges
within the images, it is preferred that the apparatus also comprises
means for detecting portions of said image indicative of edges of said
interior region of said subject, in conjunction with, preferably, means
for detecting gaps in said edges; and means for filling said gaps in
said edges, to provide a substantially continuous edge in said image.
The archiving of the eventual diagnosis of the presence or lack
of the medical abnormality is facilitated by the use of input means for
subsequent user input of data indicative of a later diagnosis of said
medical abnormality in said subject.
In one preferred embodiment, said subject is a pregnant female;
and said image is an image of a fetus. In this case, it is preferred
that said biochemical test comprises one or more tests selected from
the group consisting of: -
(i) a detection of said subject's level of human chorionic
21~1280
~ W O 94/14132 ~ PCT/GB93/02504
gonadotrophin;
(ii) a detection of said subject's level of alpha-feto protein;
(iii) a detection of said subject's level of unconjugated
oestriol;
(iv) a detection of said subject's level of pregnancy associated
placental protein-A (PAPP-A);
(v) a detection of said subject's level of the free alpha
subunit of human chorionic gonadotrophin; and
(vi) a detection of said subject's level of the free beta
subunit of human chorionic gonadotrophin.
In order to allow the detection of possible cases of, for
example, spina bifida, it is preferred that said image comprises a
sectional image of a fetal head; and said apparatus comprises means for
detecting an outline of said image of said fetal head.
In another embodiment, to allow for the detection of possible
cases of Down's syndrome, it is preferred that said image comprises a
sectional image of the nuchal fold of said fetus; and said apparatus
comprises means for detecting an outline of said image of said nuchal
fold. Alternatively, the thickness of the nuchal fold could be
ex~mined, in which case it is preferred that said image comprises a
sectional image of the nuchal fold of said fetus; and said apparatus
comprises means for detecting a thickness of said image of said nuchal
fold.
In the case of a pregnant subject, it is preferred that said
subject data comprises data indicative of the gestational age of said
fetus.
In another embodiment, a detection of possible cases of ovarian
cancer can be made. To this end, it is preferred that said image
comprises a sectional image of said subject's ovary; and said means for
detecting a quantitative measure comprises means for detecting an
outline of said image of said ovary.
Preferably said means for detecting a quantitative measure
comprises means for detecting a length of at least one axis of said
outline. More specifiçally, it is preferred that said means for
detecting a quantitative measure is operable to generate a numerical
value indicative of a ratia of lengths of axes of said outline.
A number (at least two) of quantitative measures may be detected
W O 94114132 1~8 ~ PCT/GB93102504
and the results combined using standard statistical techniques.
In respective preferred embodiments, the scanning means comprises
means for generating an ultrasound (US) image of at least said interior
region of said subject; means for generating a computer aided
tomographic (CAT) image of at least said interior region of said
subject; means for generating a single-photon emission computed
tomographic (SPECT) image of at least sai æ"lnterior region of said
subject; means for generating a positron emission tomographic (PET)
image of at least said interior region of said subject; or means for
generating a magnetic resonance (MRI) image of at least said interior
region of said subject.
The invention may be embodied as a unit which is separate from,
but works in conjunction with, a conventional ultrasound scanning
device. To this end, in a second aspect this invention provides image
processing apparatus for processing a non-invasively scanned image of
at least an interior region of a subject, said apparatus comprising:
means for detecting a quantitative measure indicative of the shape of
said image, said quantitative measure having a correlation with the
presence of a medical abnormality in said subject.
Viewed from a third aspect this invention provides a method of
non-invasive medical scanning, said method comprising the steps of:
non-invasively generating an image of at least an interior region of a
subject to be examined; and detecting a quantitative measure indicative
of the shape of said image, said quantitative measure having a
correlation with the presence-of a medical abnormality in said subject.
Viewed from a fourth aspect this invention provides a method of
image processing a non-invasively scanned image of at least an interior
region of a subject, said method comprising the steps of: detecting a
quantitative measure indicative of the shape of said image, said
quantitative measure having a correlation with the presence of a
medical abnormality in said subject.
Viewed from a fifth aspect this invention provides a scanning
method for identifying a medical abnormality comprising the steps of:
non-invasively scanning an object internally of a body to produce image
data corresponding to the object; processing the image data to generate
at least one quantitative parameter corresponding to at least one
physical characteristic of the object indicative of a medical
~ W O 94/14132 21512 8 0 PCT/GB93/02504
abnormality; generating a value as a function of the at least one
quantitative parameter, wherein said value corresponds to a
relationship between the at least one physical characteristic of the
object and the at least one physical characteristic of an object not
indicative of the medical abnormality; and determining a likelihood of
the medical abnormality based upon the value.
Viewed from a sixth aspect this invention provides a method for
identifying a medical abnormality comprising the steps of: receiving
image data corresponding to a non-invasively scanned object internal of
a body and processing the image data to generate at least one
quantitative parameter corresponding to physical characteristics of the
object indicative of a medical abnormality; generating a value as a
function of the at least one quantitative parameter, wherein said value
corresponds to a relationship between the at least one physical
characteristic of the object and the at least one physical
characteristic of an object not indicative of the medical abnormality;
and determining a likelihood of the medical abnormality from the value.
Viewed from a seventh aspect this invention provides an apparatus
for identifying a medical abnormality comprising: means receptive of
image data corresponding to a non-invasively scanned object internal of
a body for processing the image data to generate at least one
quantitative parameter corresponding to physical characteristics of the
object indicative of a medical abnormality; means for generating a
value as a function of the at least one quantitative parameter, wherein
said value corresponds to a relationship between the at least one
physical characteristic of the object and the at least one physical
characteristic of an object not indicative of the medical abnormality;
and means receptive of the value for determining a likelihood of the
medical abnormality.
Viewed from an eighth aspect this invention provides a sc~nni n~
apparatus for identifying a medical abnormality comprising: means for
- non-invasively scanning an object internally of a body to produce image
data corresponding to the object; means for processing the image data
to generate at least one quantitative parameter corresponding to at
least one physical characteristic of the object indicative of a medical
abnormality; means-for,gene,rating a value as,a function of the at least.
one quantitative parameter. wherein said value corresponds to a
W O 94114132 21$ 128 a PCT/GB93102504
relationship between the at least one physical characteristic of the
object and the at least one physical characteristic of an object not
indicative of the medical abnormality; and means receptive of the value
for determining a likelihood of the medical abnormality.
In other words, embodiments of the invention allow quantitative
abnormality detection criteria relating to `the scanned image of the
region of interest to be derived and ar~hived. These may contain
indications of the medical abnormality~i~ question, and by utilizing
digital image enhancement, detection algorithms, and pattern
recognition techniques to process the electronic signature signal train
of the scanned ultrasound (or other) image, a correlation indication
(or set of correlation coefficients) can be computed which maps or map
the scanned image into the archived criteria. That is to say, a
variable is computed that can be used to distinguish normal from
abnormal by comparison with the archived criteria.
The archived quantitative criteria can be derived from scanned
images developed using the applicable scanning technique (e.g.
Ultrasound (US), Computerized Axial Tomography(CAT), Magnetic Resonance
T~ging (MRI), etc.), which are associated with the abnormality of
interest. The derivation of the archived quantitative criteria
recognises the special image characteristics of the population having
a positive diagnosis of the abnormality, as well as the image
characteristics of an unaffected population. The archived quantitative
criteria will be appropriate for contemporary patient image generation
equipment, techniques, and for operator interface and the sCAnning
technique utilised in patient studies. In the example of an ultrasound
scanned image, quantitative criteria could include dimensional criteria
(e.g. length, width, thickness, etc.), ratio information (e.g. ratio of
length to width), form-factor information (e.g. perimeter to area),
area information, volumetric information, and relative density (e.g.
intensity) information. Criteria are developed directly from the
outputs of the scanning equipment, or from the development of
relationships incorporating these outputs, and in any case from
populations with and without the abnormality in question.
The scanned image of the region of interest (e.g. body, or
portion of body, fetus, or portion of fetus) which is being evaluated
for possible indications of the low prevalence abnormality is such that
.
~ W O 94/14132 21512 8 0 PCT/GB93/02S04
this image may contain evidence of the abnormality. That evidence may
not be readily recognized by the scanning equipment operator because of
the low prevalence of the abnormality, and the concomitant experience
level. In embodiments of the invention the scanned image is
automatically processed to emphasize the image characteristics which
describe the abnormality, provide discrimination, and relate to the
archived criteria. The automated processing of the scanned image
reduces both the amount and skill level of the operator effort
required, and by providing a reproducible representation, can reduce
uncertainty and bias which may result from operator interpretation. In
the case of an ultrasound scanned image, for example, the automated
processing can include image enhancement processes followed by pattern
recognition processes. The automatically processed data will contain
scanned image information for the specific image of interest, the
detected image describing the region being studied (e.g. body or
portion of body. fetus or portion of fetus). The automatically
processed data will then be used, in conjunction with a software
computer program tailored to the scanner type (e.g. ultrasonic), the
medical abnormality of medical interest, and the characteristic
indications as scanned, to generate image specific descriptors. These
image specific descriptors are compared with the archived quantitative
criteria which are derived from populations with and without the
abnormality.
The image specific descriptors can provide a discriminant or set
of discriminants for characterising the detected image measurement(s).
The descriptors can be combined computationally to generate one or more
figures of merit, which constitute a signature indicative of scanned
image characteristics pertinent to the abnormality of interest. The
set of image specific descriptors, combined into one or more figures of
merit unique to the detected image of interest can then be correlated
with the archived quantitative criteria developed from populations with
or without the abnormality in question.
Techniques and standards for presenting, characterising and
analysing assay data taken in conjunction with defect detection and
risk estimation studies are known (e.g. biochemical), and embodiments
of the invention are responsive to and compatible with these techniques
and standards. In other words, embodiments of the invention handle the
W O 94/14132 PCT/GB93/02504
2~ --
acquisition and processing of data to provide risk estimation. and can
relate the ultrasound risk estimation to the biochemical techniques
available for such estimations. In some embodiments, threshold or cut-
off values can be established for variables indicative of abnormality
related risk, or more formal quantitative risk estimates can be
calculated using the affected and unaffect-e~ data base, statistical
modelling, and computational package tech~qùes currently applied, for
example to biochemical risk estimation u~ing assay results, and patient
risk factor analysis.
The techniques and processes of embodiments of the invention are
independent of a specific imaging technology with respect to the
elements of detection and measurement, archiving of quantitative
criteria describing a population exhibiting positive diagnosis, and
correlation of the image specific descriptors with the archived
criteria to provide a quantitative indication of that correlation, a
risk estimate.
As an example, spina bifida, a neural tube defect, and a
congenital abnormality of the central nervous system (CNS) has an
average incidence of about one in a thousand term births, although
geographic variation in the rate of incidence does exist. The use of
contemporary high-resolution ultrasound equipment provides significant
potential for the evaluation of the fetal neural axis in early stages
of fetal development, but the identification and possible
quantification of selected CNS abnormalities such as spina bifida in
many cases require a thorough knowledge and current experience in neuro
abnormality. As currently emphasised, the accuracy and reliability of
ultrasound scanning techniques in identifying and predicting spina
bifida depend on the relevant experience and training of the operator,
the capabilities of the ultrasound scanning equipment and the scope of
the evaluation, including the time dedicated to the patient. The
sensitivity and predictive value of non-targeted initial ultrasound
examinations is probably quite low in most circumstances. Targeted
(subsequent) examinations conducted on 'at risk' patients probably have
significantly greater predictive value. In all cases involving the
detection of spina bifida using ultrasound scanning techniques,
- operator training and experience, and available examination time per
patient are limiting factors.
~ W 0 94/14132 21~12 8 0 PCT/GB93/02504
11
Embodiments of the invention provide an automated ultrasound
image processing system for detecting potential cases of spina bifida,
by the recognition and quantification of either an abnormal
configuration of the cerebellum that appears as a crescent with the
concavity pointing anteriorly ('banana sign'), or frontal bossing
('lemon sign'), or a combination of the two.
In one embodiment, for example, for the detection of the so-
called 'lemon sign', the ultrasound scanner generates a sectional image
of a fetal head and provides the means for the detection of spina
bifida by the processing of measurements taken from the cross-sectional
image. The ultrasound scanner provides means for detecting the
principal (e.g. major and minor) axes of the outline of a cross-section
of the fetal head. In this context, the major axis corresponds to the
occipital-frontal measurement and the minor axis corresponds to the
biparietal diameter of the fetal head. Various measurements can then
be made with respect to the principal axes. Preferably the principal
axes are detected by detecting the longest bisector of the outline
(this being the major axis) and then the longest axis at 90 to the
major axis (this being the minor axis). Shape characterisation
includes the identification of the principal axis of the fetal head
scan by a maximising algorithm which detects the longest bisector of
the fetal head outline, while the minor axis of the fetal head outline
is detected as the longest axis at 90 to the major axis.
Various measurements can be used to detect the condition of spina
bifida. In one preferred embodiment the apparatus comprises means for
providing the signal enhancement and processing necessary to detect the
width of the fetal head at a position of 80~ of the major axis of the
outline, and for computing a figure of merit or descriptor by dividing
that width by the sum of the lengths of the majcr and minor axes. In
another preferred embodiment the apparatus comprises means for
providing the signal enhancement and processing necessary to detect the
lengths of axes extending from the centre of the major axis at an angle
(such as 40) to the major axis, and for computing a figure of merit by
dividing the sum of those lengths by the sum of the lengths of the
major and minor axes.
B~EF DESCRIPTION OF THE DR~WINGS - -
The invention will now be described by way of example with
W 0 94/14132 21S 128 ~ PCT/GB93/02504
12
reference to the accompanying drawings, throughout which like parts are
referred to by like references, and in which:
Figure l is a schematic diagram of an ultrasound scanning and
defect detection apparatus; -~
Figure 2 illustrates various cross-sectional planes through a
fetal head;
Figure 3 represents an ultrasound scan of the head of a fetus not
suffering from spina bifida;
Figure 4 represents the ultrasound scan of a fetus suffering from
spina bifida;
Figures 5 and 6 are schematic diagrams showing measurements made
with respect to the principal axis of a fetal head scan to detect
possible fetal abnormalities;
Figure 7 is a schematic diagram illustrating the statistical
distribution of a quantitative value derived from the shape of a fetal
head outline;
Figure 8 illustrates the derivation of a likelihood ratio;
Figure 9 illustrates a risk management system;
Figures lO to 13 illustrate patient management protocols; and
Figures 14 and 15 are schematic diagrams of two further
embodiments of an outline detector for detecting an outline of a non-
invasively scanned image.
~ES~RIPTION OF THE PREFERRED EM~ODIMENTS
Figure l is a block diagram of a defect detection system which is
used as part of or in conjunction with an ultrasound scanner to detect
a spina bifida abnormality. A central processor with keyboard entry lO
and a central processor memory 20 combine to provide the computational,
data processing, transactional memory, and application program storage
and execution capabilities and implementation functionality necessary
to process digital image data, develop shape characterization
descriptors or figures of merit for specific images of interest,
perform correlation against quantitative archived criteria, and provide
a quantitative risk or categorical indication in comparison with a
specified cut-off level.
An ultrasound scanner 30 suitable for ultrasonography for
obstetric and gynaecological use will provide image data for the body
~ W O 94/14132 215 12 8 0 PCT1GB93102S04
13
or portion of a body, fetus or portion of a fetus of interest. In a
preferred embodiment for the detection of fetal abnormalities, the
image data will characterize the fetal skull, or more specifically, a
cross section of a fetal skull. The ultrasound scanner 30 may provide
the image data directly in digital form, or may require the use of a
commercially available digital scan converter 40 to convert the
ultrasound scanner analog image data to digital form, in the case of
ultrasound scanners providing analog image data only.
A high frequency emphasis filter 50 is used to emphasize the
skull (outline) portion of the image while suppressing any random noise
or "clutter" in the image. The skull portion of the image comprises
relatively narrow, sharp regions that contain higher spatial
frequencies. In contrast, the background clutter tends to be broad,
slowly rhAnging areas of the image that contain lower spatial
frequencies. The processing of the image with a high frequency
emphasis digital filtering process before applying a grey level
intensity threshold processor 60 can be used to further improve and
isolate the skull image.
The generation of the skull outline from the isolated skull image
is a two-step process requiring edge detection and gradient and zero
crossing detector 70. First, the inner and outer edges of the skull
image are sharpened by the use of a gradient operator. For example,
the so-called "Robert's cross" operator is a four point gradient
operator that generates the gradient at each pixel using the intensity
of the pixel and three neighbouring pixels. The mid-point of each edge
can be detected by applying the gradient detector a second time and
identifying zero crossings.
The second stage of the outline process is carried out by a
thinning and contour tracing processor 80. The thinning algorithm
produces a single-pixel wide trace between the two skull edges.
However, this trace may have small, spurious branches or voids that are
artifacts from the original image. The spurious branches and artifacts
are removed by an artifact processor 90 using contour tracing
- algorithms that detect the spurs by evaluating the neighbours of each
pixel on the outline trace. Voids are filled by a smoothing processor
lOO.by interpolating between isolated segments along the outline. If
.. . . . . . . .
the processing algorithms were not fully successful in removing
W 0 94/14132 PCT/GB93/02504
~ Z ~ ~ 14
clutter, the interpolation process may confuse clutter artifacts with
valid skull outline segments. The resultant skull outline can be
superimposed onto the original image to allow the operator to evaluate
the adequacy of the outline tracing process using a correction
processor llO. If necessary, the operator can "help" the algorithms by
designating which segments are skull and which are clutter using a
cursor. The interpolation process can be~ r~run with this additional
information to produce a satisfactory re~.ult.
A shape characterization processo~ 120 combines the processed
measurements extracted from the image using the algorithms developed
for that purpose. A correlation processor 130 compares the shape
characterization descriptors (e.g. figure of merit) with the archived
guantitative criteria which characterize the fetal abnormality of
interest to develop a quantitative risk indication (e.g. estimate,
relationship to threshold value).
A display l~O and archived memory 150 (to store data representing
images from an overall population, which may represent previous images
scanned by the present apparatus) are included in the system.
In each case, length, as used in computation of risk estimate,
represents the image presented by the scanner, as enhanced and
processed in accordance with detection augmentation algorithms.
Detection algorithms are permanently programmed in the system software
and associated processor.
Detection Algorithms:
a. Normal ray ratio.
b. Oblique ray ratio from BPD from centre.
c. Scallop detector.
d. Frontal radius evaluator.
e. Radius of curvature distribution.
f. Radius of curvature first derivative distribution.
g. Deviation from an ellipse.
h. Correlation with known good/bad outlines.
i. Asymmetry.
Various of the image processing techniques are described in the
book "F~lnd~mc~ntals of Digital Image Processing" (A.K. Jain, Prentice-
Hall International, 1989).
W O 94/14~'t2 ~ 1 ~ 12 8 0 PCT/GB93/02504
Figure 2, provided for purposes of illustration, indicates three
common scanning planes used in ultrasonography of the fetal skull. The
present embodiment uses a standard scanning procedure. The three
scanning planes illustrated in Figure 2 are the so-called "lateral
ventricles" plane, the so-called "bpd-hc" plane and the so called
"cerebellum" plane.
Figure 3 represents an ultrasound scan 160 of the head of a fetus
suffering from spina bifida. The scan represents a cross section
through the fetal head in the bpd-hc plane illustrated in Figure 2.
Figure 4 represents a similar scan 180 of the head of a fetus not
suffering from spina bifida. The significant differences between the
scan 160 and the scan 180, which differences are indicative of the
spina bifida abnormality, are the slight depressions 170 in the spina
bifida scan. These depressions 170 give the head the characteristic
"lemon" shape and so are known as the "lemon" sign.
In the present embodiment used to detect spina bifida,
measurements of length are enhanced, processed, and used to compute
ratios (e.g. figures of merit) which form a descriptor set for the
fetal image of interest. This descriptor set is correlated with the
archived quantitative criteria derived from a statistically valid
population having a positive diagnosis of the abnormality, spina
bifida.
In this preferred embodiment used for the detection of spina
bifida, and also applicable in instances where precise, repeatable
fetal skull cross sectional measurements must be developed from the
"raw" (e.g. unprocessed) scanner signal stream, digital image
enhancement and pattern recognition are employed. The system
processor, and its associated software control these functions.
Figures 5 and 6 are schematic diagrams showing examples of
measurements which may be made with respect to the principal axis or
axes of the fetal head outline, which measurements may then be used by
the correlation processor to detect a possible fetal abnormality. In
Figures 5 and 6 the length of the major axis of the fetal head outline
is denoted by "1", and the length of the minor axis of the fetal head
outline is denoted by "h".
Referring to the annotations on Figure 5, a variable :n
representing the shape of the fetal head outline is calculated as
W 0 94/14~2 21~ 128 ~ PCT/GB93/02504 ~
16
follows:
dl + d2
I + h
In the case of Figure 6, a similar variable m representing the
fetal head shape is calculated as follows:
s ~ s
m(~) = ' 2
I +h
The variable n and m are examples of descriptors, or figures of
merit. They may simply be displayed on a display means, compared with
various numerical ranges as described below, or used to calculate a
l;kelihood ratio or absolute risk (see below).
A thresholding or cut-off technique may be used to indicate a
potential abnormality. In this case, the shape detector compares
(correlates) the numerical value (descriptor) of the variables n or m
with a cut-off (threshold) value. If the necessary value is below or
above (as appropriate) the cut-off value, an alarm indication (e.g. a
light, a sounder or an alphanumeric message on a display) can be
automatically generated. The cut-off value is selected to reflect a
balance between the detection of the greatest possible of possible
instances of spina bifida (maximising the detection rate) against the
need to avoid an excessive false positive rate. Cut-off values
determined in trials of the embodiment are listed below:
Variable Cut-Off Value Detection Rate False Positive
(Descriptor) % Rate
dl + d2 <0.275 32(23/71)0(0/20)
l + h <0.290 55(39/71)10(2/20
sl + S2 (0=35) <o.458 65(46/71) o(o/2o)
l + h <o.466 78(55/71)10(2/20)
s~ + S2 (0=40) . ~o,4.45 . 59(42/71). . 0(0/20) .
1 + h <o.448 68(48/71)10(2/20)
~ W O 94tl4132 21512 8 0 PCT/GB93102504
17
The automated generation of the skull outline from the scanner
image can reduce the amount and skill level of the labour required
while producing a more faithful representation of the skull outline by
- reducing the uncertainty and bias that results from an operator drawn
outline. These automated processes consist of image enhancement
processes followed by the pattern recognition processes.
While digital processing of a single image does not increase the
actual information content of the image, it facilitates the pattern
recognition process by increasing the image contrast and reducing noise
and clutter. Because of the sharp differential in the relative
propagation velocities of sound through bone and tissue, the skull
portion of the fetal image tends to be higher intensity than the other
(tissue) areas of the image. This characteristic allows the
application of a pixel intensity threshold process to the digitized
image to delete clutter while retaining the skull data.
A second characteristic of the skull image permits the use of
high frequency emphasis filtering to emphasize the skull portion of
image while suppressing the clutter in the image. The skull portion of
the image consists of relatively narrow, sharp regions that contain
higher spatial frequencies. The background clutter tends to be broad,
slowly rh~nging areas of the image that contain lower spatial
frequencies. The processing of the image with a high frequency
emphasis digital filtering process before applying the intensity
threshold process is used to further improve and isolate the skull
image.
The generation of the skull outline from the isolated skull image
is a two step process. First, the inner and outer edges of the skull
image are sharpened by the use of a gradient operator. The Robert's
cross operator is a four point gradient operator that generates the
gradient at each pixel using the intensity of the pixel and three of
its neighbours. The mid-point of each edge if detected by applying the
gradient detector a second time and identifying zero crossings.
The second stage of the outline process is a thinning and contour
- tracing process. The thinning algorithm produces a single-pixel wide
trace between the two skull edges. However, this trace may have small,
spurious branches or voids that are artifacts from the original image
The spurious branches are removed using contour tracing algorithms that
-
W 0 94/14132 ~ 2 ~ ~ ~ 2 8 ~ PCT/GB93/02504 ~
18
detect the spurs by evaluating the neighbours of each pixel on the
outline trace. Voids are filled by interpolating between isolated
segments along the outline. If the processing algorithms were not
fully successful in removing clutter, the interpolation process may
confuse clutter artifacts with valid skull outline segments. The
resultant skull outline can be superimpo~d onto the original image to
allow the operator to evaluate the adequacy of the outline tracing
process. If necessary, the operat,Q~ can "help" the algorithms by
designating which segments are sku~l and which are clutter using a
cursor. The interpolation process can be rerun with this additional
information to produce an optimized result.
Preferably the results of two separate ultrasonography tests are
combined, in order to increase the reliability of the detection of the
possible medical abnormality. To this end it is preferred that the
embodiment detects at least two quantitative measures relating to the
image; and that the variable (descriptor) indicative of the shape of
the image is detected in response to the at least two detected
quantitative measures. Increased dimensionality of the correlation,
and higher orders of discrimination will be provided where required by
detection strategy.
In this embodiment, and in order that an operator is alerted to
the detection of a possible medical abnormality and can then (if
appropriate) initiate further testing or investigation of the patient,
it is preferred that the embodiment comprises means for automatically
generating an alarm indication in response to a detection that the
variable (descriptor variable) lies in the range indicative of a
possible medical abnormality (e.g below the thresholds described
above).
This is one of a range of possible strategies for conducting,
intervening, utilizing, and reporting on data presented by the
embodiment. Other strategies will now be discussed.
The alarm indication could take the form of, for example, an
audible warning or the illumination of an indicator light on an
operator's console. In an alternative embodiment an alphanumeric
display is used for use alone or in combination with other indicators
of the condition under investigation, to display an appropriate alarm
.
message.
W 0 94/14132 ~ ~ S 12 8 0 P~T/GB93/02504
19
The acquisition, processing. and archiving of ultrasound defect
detection data, and the development of risk estimates will be
consistent with the techniques and standards employed in risk
estimation programs (e.g. biochemical).
Another embodiment allows for the detection of potential cases of
Down's syndrome. The overall birth prevalence of Down's syndrome
(trisomy 21) is about l.3 in lOOO births, and is the most common cause
of severe mental retardation in humans. Performing a fetal karyotype
is the definitive method of diagnosis, and biochemical assays can be
used in combination to identify a high risk group and provide for
individual pregnancies. While many of the structural abnormalities
associated with Down's syndrome are too subtle to be detected with
current ultrasound equipment, several morphologic signs have been
identified which indicate an elevated risk of Down's syndrome, and
which can be detected by contemporary ultrasound techniques. These
include a thickened nuchal fold, shortened femurs, and hypoplasia of
the middle phalanx of the fifth digit.
A modified transverse view of the fetal head, including the
cerebellum and occipital bone is used to detect abnormal thick~ning of
the soft tissues at the back of the fetal occipital. The nuchal fold
is usually measured from the outer edge of occipital bone to the outer
edge of the fetal skin. As in the case of spina bifida, this
embodiment provides the automatic signal processing, image enhancement,
and pattern recognition to detect potential cases of Down's syndrome
risk, as indicated by a thickened nuchal fold. In other words, the
outline or thickness of the nuchal fold is detected.
Fetuses exhibiting Down's syndrome tend to have somewhat
shortened femur lengths, compared to an unaffected population. The
femur length:biparietal diameter (BPD) ratio test also tends to be
reduced compared to unaffected populations. The techniques and
implementation provided by the spina bifida embodiment described above
- can be used to interpret femur length and the ratio of femur length to
BPD.
In the detection of ovarian cancer (or carcinoma of the ovary,
CaO), the use of transvaginal sonography provides a method of screening
- for ovarian cancer. Ovarian cancer, which constitutes a significant
proportion of all gynaecological malignancies, and which has a poor 5
-
W O 94/14132 21 S 12 8 ~ PCTIGB93/02504
year survival rate, is often first detected in an advanced stage. The
detection of ovarian cancer in an earlier stage may improve the
prognosis. A further embodiment relating to transvaginal ultrasound
images of the ovaries, will refine the capability of diagnostic
techniques to recognize, for example, relatively enlargement of the
ovary or abnormality in shape and morphdlogical pattern.
The embodiment for the detec~on of ovarian cancer related
abnormalities employs intensity (density), volumetric and dimensional
(shape), and computed mass and mass differential descriptors and
discriminants.
Figure 7 is a schematic diagram illustrating the relative
distribution of exemplary populations affected and unaffected by a
medical abnormality. In particular, Figure 7 is used here to
illustrate the methodology be which risk estimates (e.g. likelihood)
that a given condition, such as a fetal defect, are calculated. The
methodology applies to the ultrasound spina bifida detection technique,
to biochemical marker tests, and to the application of the present
techniques to ultrasound and other imaging techniques which examine the
body and portions of the body, the fetus, and portions of the fetus for
indications of risk presented by other abnormal conditions. Figure 7
shows the relative frequency distributions of a variable (e.g. n, m)
among an affected 510 and unaffected population 500, and is plotted in
the instance in multiples of the median for the unaffected population.
This illustrative example is derived from gestational age corrected
maternal serum alpha-fetoprotein data for open spina bifida. the
presentation, treatment, and statistical techniques used for this
biochemical marker are well described and are typical of the
methodology employed in the computation of likelihood estimates (risk).
The derivations can allow for the influence of such factors as
gestational age, maternal age, maternal weight, race, geographic
residence, and insulin dependent diabetes.
The application of population statistics for a fetal defect of
interest, and the calculation of likelihood ratios utilize well
established statistical techniques. The embodiments incorporates these
established statlstical and computational techniques. The embodiments
may also include the incorporation of features responsive to clinical
.
practice requirements, patient safety, and Food and Drug
W O 9411413~ Z1 PCT/GB93/0~504
A~mini.~tration, and/or European Community criteria.
The apparatus can receive patient-related data, such as the
patient's age or weight, or the gestational age of a fetus, to reduce
the effects of random noise on the distributions. Generally this will
narrow each of the two distributions shown in Figure 7.
- Figure 8 illustrates the calculation of a likelihood ratio.
Given the statistical distributions illustrated in Figure 7, the
likelihood ratio for a sample 505 is the ratio of the heights of the
two relative distribution curves at that point, i.e. f2/fl. The
likPlih~od ratio represents how much more likely that subject is to
have the abnormality than a general member of the population, and so
can be converted to an absolute risk of the abnormality by multiplying
the likelihood ratio by the "prior risk" exhibited by the whole
population. The likelihood ratio and absolute risks can be displayed,
thresholded or used to trigger alarms as described above. The two
curves shown in Figures 7 and 8 can simply be stored in the memory 150
by storing their means and standard deviations. If more variables are
being considered, correlation coefficients between the variables should
also be stored.
Alternatively, if a thresholding technique is used, the overlap
of the two distributions means that a proportion of the affected
population 500 will fall below the threshold and a proportion of the
unaffected population 510 will fall above the threshold. These
proportions constitute false negative results and false positive
results. In order to reduce these false positives and negatives,
additional factors such as the maternal weight may be considered to
adjust the threshold values.
Figure 9 is a schematic diagram illustrating a further embodiment
which recognises the common nature of much of the analytical and
computational treatments of ultrasound and biochemical risk estimation
data. The principles of risk estimation based on univariate and
multivariate (e.g Gaussian) distributions are common to the ultrasound
defect detection, and the more mature, biochemical testing and fetal
defect risk estimation techniques.
Biochemical risk estimation using the results of assays involving
two or more markers is performed by the application of software
computer programs to interpret results. These programs take account of
21S12~0
W O 94/14132^ PCTJGB93/02504
22
disease prevalence distributions of the markers and their correlation,
and other factors that may be relevant such as gestational age and
maternal weight. Ultrasound defect detection and risk estimation use
these automated computer techniques, as well as the image associated
techniques described in this disclosure. In both the biochemical case
and the ultrasound case, "raw" (unprocessèdp data will be available
either on a "real time" basis, using an' ;~RS 232" or other standard
communications protocol, or on a delaye`d"time basis ldays, weeks), by
data entry.
Figure 9 illustrates a risk management system having
computational elements (e.g. a central processor unit 610, a memory
620, an input-output processor 630, and a local/wide area communication
network 640) which meet the requirements of both ultrasound and
biochemical derived risk estimation methodologies are consistent and
compatible. The risk management system 600 provides the functional
means of implementing both ultrasound and biochemical risk estimation
using data sources now available. This embodiment integrates the
elements of ultrasound detection with biochemical and patient data in
a single, integrated risk management central processing unit.
Figure lO illustrates a decision making protocol which
incorporates data derived by biochemical and ultrasound techniques to
detect congenital abnormalities in the developing fetus. A level l
ultrasound scanned image, using standard obstetric/gynaecological
ultrasound equipment, augmented by the defect detection processor 300
(according to an embodiment of the invention) is used for dating (e.g.,
the assessment of gestational age), detection of possible fetal
abnormalities as described in this disclosure (e.g., spina bifida,
Down's syndrome), and for any other examinations requested by the
physician and/or the laboratory. The estimation of gestational age
resulting from the ultrasound examination utilizes standard techniques
and will be derived from measurements appropriate to the specific
circumstances (e.g. reliability, ease of measurement). Measurements
may include crown-rump length (CRL) or biparietal diameter (BPD) for
assessment of gestational age. The estimated gestational age resulting
from this dating measurement can improve the performance of biochemical
screening since the biochemical marker measurements are influenced ~y.
gestational age.
~ W O 94/14132 215 12 8 0 PCT/GB93/02S04
23
The data derived from the ultrasound image augmented by the
present embodiments will provide an indication of fetal abnormalities
by processing and enhancing measurement data, calculating descriptor
variables (e.g., n and m), and relating the variable value(s) tocriteria (e.g., cut-off values). The results of the defect detection
process will be available at a time in the gestational cycle which is
influenced by the age assessment technique, and the specific ultrasound
abnormality detection technique. Abnormality detection using
ultrasound techniques can be accomplished relatively early in the
gestational cycle. allowing directed follow-up and biochemical testing
where results indicate a fetal abnormality.
Maternal serum biochemical testing 310 is based on the
measurement of such marker levels as alphafetoprotein (AFP), human
chorionic gonadotrophin (hCG), unconjugated oestriol (uE3), pregnancy
associated placental protein A (PAPP-A), free alpha subunit of human
chorionic gonadotrophin (alpha hCG), and free beta subunit of human
chorionic gonadotrophin (beta hCG) (from a pregnant patient in a
specified gestational range) in maternal serum. Risk estimation
required integration of the individual marker levels, using appropriate
statistical methods. In general the results are influenced by
estimated gestational age, so that more effective ultrasound dating
increases the reliability of the risk estimate. The present
embodiments allow the ultrasound markers (e.g. m, n) to be combined
with the biochemical tests to improve the performance of screening.
Should maternal serum biochemical testing indicate abnormal
results, testing a second specimen 320 may be carried out but is not
normally rec.- ~ded.
Should the biochemical testing of a second specimen of maternal
serum indicate abnormal results level 2 ultrasound 330, high resolution
(augmented by apparatus according to an embodiment of the invention),
such ~s the defect detection processor described hereinbefore, will be
performed. The level 2 ultrasound examination 330 will provide refined
gestational age assessment, fetal abnormality detection as provided by
these embodiments and provision for expanded scope and/or directed
follow-up. Abnormal results from the multi-stage screening approach
indicate the need for amniotic fluid testing, fetal karyotyping and a
more detailed diagnostic ultrasound examination.
WO 94/14132 PCT/GB93/02504
21S~2~
24
Amniotic fluid biochemical testing 340 is similar to maternal
serum biochemical testing 310 in its use of measured marker levels,
interpretation in relation to specified cut-off levels, integrated
statistical processing of individual marker results, and calculation of
risk estimates for defects of interest. Amniotic fluid tests include
alpha-fetoprotein and acetylcholinesterase measurement and fetal
karyotyping. Detailed ultrasonography is also carried out.
Appropriate follow-up includes patient counselling.
The present embodiments can thus provide a means for augmenting
an accepted protocol recommended (e.g. by the United States Food and
Drug Administration) for use in decision making by and counselling for
patients undergoing testing for fetal abnormalities. This protocol,
outlined in Figure 9, does not include the automatic defect detection
capabilities provided here. The sequence of the protocol steps, and
the criteria for both need and timing in the gestational cycle are
modified to improve the performance of screening for fetal defects.
In contrast, in the standard protocol illustrated in Figure 11,
the maternal biochemical serum testing 400 is identical to the maternal
serum biochemical testing 310 described in Figure 10. The second
specimen maternal serum biochemical testing 410 is identical to the
second specimen testing 320 shown in Figure 10. The biochemical
amniotic fluid testing 430, is identical to the amniotic fluid
biochemical testing 340 shown in Figure 10. Confirmatory tests 350 are
similar to these indicated 440 in Figure 10. In the modified protocol
shown in Figure 10, all ultrasound testing 300, 330, and 350, include
defect detection capability resulting from the augmentation of the
standard ultrasound scanner by these embodiments.
Protocol step 330, the early augmented ultrasound scanning test
incorporating an embodiment of the present invention, is introduced
formally into the recommended protocol to provide the benefits of early
defect detection with respect to treatment, additional testing,
decision making and patient counselling. The embodiments thus
inflùence the decisions and the timing of subsequent protocol phases.
All protocol steps in Figure 10 involving ultrasound sc~nning; 300,
330, and 350 include the augmentation provided by the embodiments, the
aùtomatic defec~ detection capability.
- In the standard protocol, Figure 11, protocol steps involving
~ W O 94/14132 21 S 12 8 0 PCT/GB93102504
ultrasound scanning, 420 and 440, utilize standard ultrasound scanning
technique, and do not provide automatic defect detection capability.
The standard protocol does not provide the flexibility and timing
benefits of the step 1 augmented ultrasound test 300 indicated in
Figure 10.
Figures 12 and 13 are schematic illustrations of two further
decision making protocols incorporating apparatus according to the
present embodiments. In Figure 12, an initial blood test and a level
1 ultrasound (as described above) are carried out in parallel, and in
Figure 13 they are carried out sequentially. A positive indication in
either case leads to more detailed testing. In either case, whether
the pregnancy leads to birth or termination, the actual outcome (i.e.
the actual presence of the abnormality) is stored in the memory 150
along with the quantitative descriptors derived from images of that
patient. These values can be used to modify the population
distributions of Figure 7, to tend to increase the accuracy of future
diagnoses.
Figures 14 and 15 are schematic diagrams of two further
embodiments of an outline detector for detecting an outline of a non-
invasively scanned image. In- each of Figures 14 and 15, the image
outline is indicated by the user tracing the outline on the screen of
a video display device 970. In the embodiment shown in Figure 14, the
user employs a light pen 980, connected to an image processor 950 and
a video memory 960 storing the image data, to trace the image outline,
whereas in Figure 15 the user employs a mouse-or key-driven cursor
generated by an image processor 990 to trace the image outline.
The embodiments provide a positive means for modifying the
generally accepted decision making protocol for prenatal detection of
Down's syndrome and neural tube defects to provide benefits of timing,
flexibility, and opportunity for treatment and additional early
testing.
Although illustrative embodiments of the invention have been
described in detail herein with reference to the accompanying drawings,
~ it is to be understood that the invention is not limited to those
precise embodiments. and that various changes and modifications can be
effected therein by one skilled in the art without departing from t-he
scope and spirit of the invention as defined by the appended claims.
