A Preliminary Approach for the Segmentation of Mitral Valve

Regurgitation Jet in Doppler Ecocardiography Images

Eva Costa

, Nelson Martins

1,2

, Malik Saad Sultan

2,3

, Diana Veiga

, Manuel Jo

ao Ferreira

1,4

Sandra Mattos

and Miguel Coimbra

2,3

Enermeter, Sistemas de Medic¸

ao, Lda, Braga, Portugal

Faculdade de Ci

encias, Universidade do Porto, Portugal

Instituto de Telecomunicac¸

oes, Porto, Portugal

Centro Algoritmi, University of Minho, Guimar

aes, Portugal

ırculo do Corac¸

ao de Pernambuco, Recife PE, Brazil

Keywords:

Mitral Regurgitation, Echocardiography, Segmentation, Doppler, Heart, Medical Image Processing.

Abstract:

Rheumatic Fever and Rheumatic Heart Disease remain a major burden among children in developing countries.

Echocardiography with colour ﬂow Doppler is key to early diagnosis. However, the technique requires time

and experienced operators, which are scarce resources in the affected areas. Automatic segmentation of colour

Doppler regurgitation jets could, potentially, reduce the cost of screening, and spread diagnostic accessibility

for a larger number of patients. Ultrasound processing is very challenging due to speckle noise and similarity

of representation of all kinds of tissue. Region-based active contours are suitable tools for the segmentation

in cases of intensity heterogeneities, which makes them interesting algorithms for left atrium segmentation.

HSV colour space describes colour in terms of hues and saturation, which may facilitate the translation of

medical interpretation of the Doppler pseudo-colour into mathematical expression for colour segmentation. A

total of 979 frames from 20 sequences were manually annotated and used to validate the proposed pipeline.

Overall, the results for colour pattern segmentation are promising (sensitivity=0.91 false detection rate=0.10),

but further developments are required for the atrium segmentation (sensitivity=0.80, false detection rate=0.28).

1 INTRODUCTION

Rheumatic Heart Disease (RHD) is the chronic,

longterm sequel of repeated untreated Acute

Rheumatic Fever (ARF) episodes (E. Marijon, 2009).

ARF results from an autoimmune hyper response to

untreated streptococcal infections of the pharynx.

Most of the patients endure active heart inﬂammation

(carditis) for several months, with the vast majority

having valvulitis on the left side of the heart. The

most affected structure is the mitral valve complex,

which is essential to ensure the unidirectional sys-

temic blood ﬂow from the left atrium to the left

ventricle. Structural changes, such as thickening

or retraction of the valve and its valvar apparatus,

are often found in RHD patients. This may cause

severe strain to the heart leading to abnormal function

with prolapse, stenosis and/or regurgitation. The

major burden of ARF is found among children (5-16

years old) from low to middle income countries and

indigenous populations, and it is directly linked to

hygiene practices and limited access to antibiotics

and medical care (M. Gewitz, 2015).

Most cases of RHD are only diagnosed after

symptoms of heart failure arise and without a clinical

history of ARF. Late diagnosis often results in short

life expectancy, leading to an estimation of 233,000

deaths per year worldwide. Unlike ARF, advanced

cases of RHD are not treatable via secondary prophy-

laxis, since irreversible damage of the heart tissue is

often present (M. Essop, 2014). Surgery is required

for these patients, many of which undergo replace-

ment of their native valves. Early detection increases

life expectancy signiﬁcantly. Screening campaigns

have shown that most of the early valvar compromise

is silent, thus amenable to go undetected on clinical

examination alone. Borderline or subclinical RHD

cases may not present classical symptoms, but only

mild alterations that can only be identiﬁed if subject

to echocardiography (B. Remenyi, 2012).

Even though echocardiography is one of the most

cost effective solutions to screen for heart valve

Costa E., Martins N., Sultan M., Veiga D., Ferreira M., Mattos S. and Coimbra M.

A Preliminary Approach for the Segmentation of Mitral Valve Regurgitation Jet in Doppler Ecocardiography Images.

DOI: 10.5220/0006248900470054

In Proceedings of the 10th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2017), pages 47-54

ISBN: 978-989-758-215-8

anomalies, the average socioeconomic context of the

affected populations is not favourable to this kind

of monitoring RHD patients need. To compensate

for the lack of resources, several screening studies

have been made in different regions to assess the

practicability of some cost reducing measures. In

2015 Lu et al. (J. Lu, 2015) studied the possibility

of simplifying the criteria for RHD detection with

no substantial losses in sensitivity and speciﬁcity.

Several studies reported on the feasibility of RHD

screenings being done by non-expert health workers

such as nurses, after being subject to brief training.

Results are favourable and point to the potential use

of the technique in low-resources settings. However,

erroneous regurgitation jet identiﬁcation and incor-

rect measurement are the predominant causes of false

positives, as the correct and consistent interpretation

of these ﬁndings requires experience (M. Ploutz,

2015) (M. Mirabel, 2016) (D. Engelman, 2016).

Moreover, the manual annotation of the regurgitation

jets is time consuming and is highly operator depen-

dent and subjective (Roelandt, 1986) (R. Castello,

1992). The Automatic detection of regurgitation jets

would signiﬁcantly decrease time and experience de-

pendency of the user. Mitral Regurgitation is visually

assessed via parasternal long axis (PLAX) and apical

4 chamber (A4C) views, with colour ﬂow Doppler

(Figure 1). The pseudo-colour is superimposed on

the brightness mode, representing the movement of

the blood cells in the heart. The Hue represents the

direction of the ﬂow, where the blue and red hues

indicate, respectively, that the ﬂow is moving away

and towards the transducer. The saturation represents

the magnitude of the velocity. Regurgitation jets are

often represented in plain blue hues or high saturation

colour mosaics (blue, yellow and orange hues) due to

turbulence, which may or may not be involved in a

blue or red envelope.

In this paper we propose a semi-automatic

method for the detection of mitral regurgitation jets

candidates, based on HSV space for colour segmen-

tation and localizing active contours for anatomical

segmentation. The paper is organized as follows:

section 2 describes the materials and methods used

for this paper; section 3 presents the proposed method

for regurgitation jet detection; section 4 presents the

results and discussion for the methods exposed in

sections 3; section 5 concludes this paper.

Figure 1: PLAX (left) and A4C (right). Anatomical scheme

(top) and colour ﬂow Doppler mode (bottom).

2 MATERIALS AND METHODS

2.1 Dataset

The dataset used, in this study, consists on 20 se-

quences of paediatric echocardiography, forming a to-

tal of 979 frames, all stored in DICOM format. The

sequences were acquired during the Heart Caravan of

2016, a health care provision initiative which took

place in the State of Para

ıba organized by the non-

governmental organization C

ırculo do Corac¸

ao. Pa-

tients were all children, with mean age of 10 years

and a standard deviation of 4 years. Half of the pa-

tients were female and the other half were male. All

the patients presented some degree of mitral regurgi-

tation and 5 of them had a previously stablished di-

agnosis of Rheumatic Fever. Eleven of the sequences

were acquired with Vivid I, portable echocardiogra-

phy equipment from GE, with paediatric cardiovas-

cular probe 6S-RS with range of frequency 2.7 to

8 MHz, and image resolution of 422 x 636 pixels.

The frames are stored uncompressed, via Explicit VR

Little-endian transfer syntax. The other 9 were ac-

quired with CX-50, portable echocardiography equip-

ment from Philips, with the paediatric probe S5-1

with frequency range of 1 to 5 MHz, and image reso-

lution of 600x800 pixels. The frames are compressed

in lossless RLE format. Fifteen of the sequences were

acquired via parasternal long axis view, and the other

5 via apical four-chamber view.

2.2 Manual Annotations

The manual annotations used as ground truth for the

results evaluation were drawn by a research associate

experienced in echocardiography analysis. Three cat-

egories of objects were manually segmented: the

BIOIMAGING 2017 - 4th International Conference on Bioimaging

left atrium boundaries, the regurgitation jets and the

colour pattern candidates. The colour pattern candi-

dates include detections on the diastole phase, which

cannot be related to regurgitation, and detections on

the systole phase, which may or may not be related to

regurgitation.

2.3 Performance Metrics

In the case of mitral regurgitation jets, it is fundamen-

tal to have a measure of the portion of jets which are

detected. Sensitivity, or true positive rate (1) provide

that information:

T PR = T P/(T P +FN) (1)

On the other hand, it is also important to know

if the detection of negatives is achieved. Speciﬁcity

or true negative rate (2) gives the proportion of well

detected negatives for all the real negatives:

T NR = T N/(T N + FP) (2)

Considering the size of the images and the

dimensions of the objects of interest, the majority

of the image will be composed of negative samples,

very easy to detect (note that a signiﬁcant part of the

image has no information). Therefore, speciﬁcity

will not be able to measure small variations of jets

detection. An alternative to the speciﬁcity is the false

detection rate (3) which gives the proportion of the

false positives in all the detections:

FDR = FP/(FP + T P) (3)

We will use the same metrics for the segmenta-

tion of the left atrium, evaluating the pixels inside or

outside of the contour, since the purpose of the seg-

mentation is to deﬁne a region of interest and not to

get a precise boundary of the chamber.

3 REGURGITATION JET

DETECTION

The detection of regurgitation jets in colour ﬂow

Doppler images requires two phases: colour segmen-

tation and anatomical segmentation. Colour segmen-

tation aims for the detection of the colour patterns

of interest, while anatomical segmentation aims to

deﬁne the region of search for those patterns. The

method we propose is represented in Figure 2. Colour

segmentation is divided in two phases: pseudo-colour

isolation and colour pattern segmentation. The result

is a set of candidates of jet regurgitation. The result

of the anatomical segmentation is a set of masks of

the left atrium, which conﬁne the search for colour

patterns.

Both segmentation phases will be explained

throughout this section.

Figure 2: General scheme of the proposed method.

3.1 Colour Segmentation

Based on the clinical assumption that mitral regur-

gitation is always represented by either blue hues

(blood cells move away from the ultrasound trans-

ducer) or high intensity colour mosaics (yellow and

orange hues), we propose a colour-based approach for

the segmentation of regurgitation pattern candidates.

The pseudo-colour mapping of Doppler ﬂow follows

a standard of what is known to be natural and intu-

itive in terms of the human eye perception. The hu-

man perception of colour is often described based on

hues and saturation. We choose to work on the HSV

(hue-saturation-value) colour space, because it offers

the possibility of formulating a mathematical model,

based on physicians interpretations of Doppler colour.

Hue and Saturation together form the chromaticity,

which is the information of colour. Hue describes

a pure colour and saturation describes its purity (a

measure of the dilution of the colour by white light)

(R. Gonzalez, 2008).

3.1.1 Pseudo-colour Isolation

Saturation channel is typically used to isolate regions

of interest in the hue channel. Pseudo-colour has

inherent high degrees of purity, which translate into

high saturation (R. Gonzalez, 2008). Therefore, it is

known that the colour information will be present in

pixels of high saturation. A mask can be generated

by thresholding the saturation channel, to isolate the

pseudo-colour from the grayscale tissue background.

The area of the mask is taken as a measure of its

completeness. A set of thresholds were tested, each

one being a percentage between 0% and 100% of the

maximum intensity of the Saturation channel, with a

step of 10%. The test was made for 15 random sam-

ples from the dataset (3 frames from 5 videos) and

the results are represented in Figure 3. For thresholds

near 0% the area is maximum and for thresholds near

A Preliminary Approach for the Segmentation of Mitral Valve Regurgitation Jet in Doppler Ecocardiography Images

100% the area is minimum. In the middle values it

is expected to ﬁnd a range of thresholds which will

lead to similar area values, since the saturation inten-

sities will not vary signiﬁcantly for the pseudo-colour

pixels. Tests conﬁrm the existence of a stabilization

plateau, and so, the value in the centre was used as a

threshold, which is 50%. An example of the result of

colour isolation is represented in Figure 4.

Figure 3: Total mask area for each threshold. Each line

represents one frame from 5 randomly selected sequences.

Figure 4: Original PLAX view frame (left) and result from

pseudo-colour isolation (right).

3.1.2 Colour Pattern Segmentation

As previously stated, pseudo-colour formation in

Doppler images relies on the basic convention that

reddish and blueish hues represent blood moving to-

wards and away of the probe. When accessing the

blood ﬂow through the mitral valve, normal unidirec-

tional ﬂow from the left atrium to the left ventricle

should be represented by reddish hues. Other hues

are a product of abnormal motion of blood, such as

regurgitation (Figure 5) or turbulence.

We propose a colour segmentation based, on the

hues, to get candidates of mitral regurgitation. As-

suming that we are interested in all hues but reds and

dark oranges, a simple threshold of the Hue channel

may be sufﬁcient to remove the red component of the

pseudo-colour. Yellow and red hues are represented

by low magnitudes, whereas blue hues are represented

by high magnitudes. The threshold for the red hues

removal is roughly deﬁned, during acquisition time,

by visual inspection of the colour bar provided by the

equipment (Figure 6). The value is set to 10% of the

maximum magnitude of the Hue channel.

Since the pseudo-colour in Doppler only repre-

Figure 5: Regurgitation jets in A4C view. Regurgitation is

represented in a colour mosaic (left) or in blue hues (right).

Figure 6: Colour-bar from Vivid I, GE and Hue channel

magnitude for each colour.

sents one direction of the blood motion, it is expected

to have discontinuities in the representation of jets.

Physicians will often extrapolate the segmentation of

the regurgitation jets to include these regions of pre-

dicted regurgitation. The segmentation is reﬁned us-

ing morphological methods to cover some of the dis-

continuities: morphological closing is applied to con-

nect close blobs and ﬂood ﬁll is applied to ﬁll holes.

3.2 Anatomical Segmentation

World Health Organization guidelines for the

echocardiographic assessment of mitral valve abnor-

malities indicates the use of the A4C and PLAX win-

dows in the diagnosis criteria extraction. Regurgita-

tion jets should be found within the left atrium region

(B. Remenyi, 2012). Both views include other struc-

tures of the heart besides the left atrium and the mitral

valve complex. When selecting a region of interest on

the image for visualization of the colour ﬂow, techni-

cians often include neighbour areas such as part of

the left ventricle, the aorta valve or the right atrium.

These areas, will also have a Doppler response but,

for the assessment of the mitral valve, they are of no

use. Because of that we decided to restrict the search

zone, so that only the left atrium is analysed.

Automatic segmentation of anatomical structures

in ultrasound images is very challenging, due to

the low spatiotemporal resolution, inherent acoustic

interferences and speckle noise, and lack of intensity

and texture differentiation between different tissues

(S. Mazaheri, 2015). Active contour models are very

popular approaches for medical image segmentation.

They usually belong to one of two categories: edge-

based or region-based. Edge-based active contour

models have good performance segmenting regions

BIOIMAGING 2017 - 4th International Conference on Bioimaging

of heterogeneous intensity, but are highly sensitive

to noise. Region-based are usually more robust to

noise and initialization, but they assume that the

intensities of each region to be constant, which

makes them not suitable for heterogeneous cases

which we often ﬁnd in ultrasound images. Localizing

region-based active contours are proposed with the

objective of allying the advantages of both categories

while suppressing their drawbacks. Foreground

and background can be described in terms of small

regions; thus, heterogeneity of the image is not a

problem. A disk kernel (B) moves along each point

of the initial contour, computing exterior and interior

energies, as represented in Figure 7. The total energy

is given by:

E(φ) =

Ω

δφ(x)

Ω

B(x,y).F(I(y),φ(y))dydx

+λ

Ω

δφ(x)



∆φ(x)



(4)

Where I denotes the input image at the domain Ω,

δφ(x) is the portion of contour centred at x and B is the

neighbourhood kernel around the point (x,y), which is

used to deﬁne the localizing area. These elements are

graphically represented by the red circle (B) and the

yellow point (x,y) in Figure 7. F is the internal energy

function. The last term refers to the continuity of the

contour line and is scaled by a factor λ.

An energy optimization algorithm will move the

point by ﬁtting a model to the region.

Figure 7: Graphical representation of the localizing active

contour method. The red circle is the neighbourhood, B, of

the yellow point (x,y) (S. Lankton, 2008).

For this work, the internal energy function F used

was the Uniform Model, deﬁned as:

F = H(φ(y))(I(y) −u

)

+ (1 − Hφ(y))(I(y)− v

)

(5)

Where H(φ(y)) represents the exterior of the contour

in the neighbourhood (B) and u

represents its mean.

While (1-Hφ(y)) represents the interior and v

repre-

sent its mean (S. Lankton, 2008).

For the left atrium segmentation in 2D+t echocar-

diographic sequences, we proposed the application of

localizing region-based active contours. The Doppler

pseudo-colour that is superimposed in the brightness

mode image will interfere with the active contour ad-

justment. For that reason, we use the mask obtained at

the pseudo-colour isolation step, in order to get an es-

timation of the grayscale brightness mode at the back-

ground. The pixels inside the colour mask are passed

to the active contours algorithm as low random val-

ues (below the average of intensity of the image), to

mimic the typical non-tissue speckle affected aspect

of the interior of the atrium. This assumption may

seem as a possible source of errors for the contour

adaptation, specially in cases when the colour mask

is signiﬁcantly large when compared with the atrium.

However, practical results suggest this is not a signif-

icant drawback for the method. A possible reason for

the ability of adaptation of the localizing active con-

tours in these cases is the internal force term, which

limits the deformability of the contour, maintaining

its shape. The penalizing parameter for the arc-length

of the curvature (λ) must be sufﬁciently small so it al-

lows deformation of the contour to ﬁt the corners of

the atrium, but high enough to prevent leakage. Four

values between 0.4 and 1.0 were tested and 0.6 was

the one that seemed to keep the best compromise be-

tween the requirements.

Minimal user interaction is required for contour

initialization: two points, one at the centre of the left

atrium and one at the endocardium (inner boundary

of the atrium). The initialization mask is a circle cen-

tred at the ﬁrst point with radius equal to the distance

between the two points. The stopping condition in

this method is the number of iterations. Since the re-

gional mask will move along each point of the contour

at each iteration, the process may be computationally

heavy. The number of iterations for the ﬁrst frame

was selected empirically after testing multiple values

of radius of local region and checking at which num-

ber of iterations convergence happened. Frequently

there was no further active contour adaptation or im-

provement around 100 iterations. For the following

frames, the contour initialization was given by the

previous frame contour result. Assuming that the

variation between successive frames is smaller than

the variation between the user input mask and the ﬁnal

result of the ﬁrst frame, the number of iterations for

convergence should be smaller. Similar experiment

was made for those frames, starting at 100 iterations

and reducing gradually. The ﬁnal iterations number

was 50.

Radius of the local region mask is of great rele-

vance for the performance of the algorithm. A study

was made by the author (S. Lankton, 2008) to anal-

yse the effect of the radius on the resulting contour.

The radius should be chosen considering the scale of

A Preliminary Approach for the Segmentation of Mitral Valve Regurgitation Jet in Doppler Ecocardiography Images

the object to be segmented and its distance to the sur-

rounding. Very small radius makes the internal en-

ergy function to approximate on of an edge detector

while very large radius make it tend to a global region

statistics approach. Once again, a compromise must

be made to accommodate both requirements of the lo-

cal and global features. Intermediate values of radius

usually result in similar outcomes, but with different

speeds of convergence. Taking the typical dimensions

of the left atrium on the echocardiography windows

into account, a set of eleven radii were tested from 10

to 20, with a step of 1. Results showed small varia-

tion for intermediate values, and a radius of 16 was

chosen.

Before being input to the localizing active con-

tours framework, the images are converted into

grayscale, a histogram equalization is performed and

Gaussian blur is applied for smoothing.

4 RESULTS AND DISCUSSION

4.1 Colour Pattern Segmentation

The dataset was subject to the colour isolation and

pattern segmentation pipeline. The search for patterns

was restricted to the regions of the left atrium given by

the manual annotations. The results from this process

are evaluated against the manual annotations of colour

patterns (regurgitation jets and candidates).

The mean values for the performance metrics are:

sensitivity of 0.913 and false detection rate of 0.103.

Sensitivity shows that more than 91% of the pixels

are correctly identiﬁed as jets or candidates by the

pipeline, which is a satisfactory result. An example

can be seen in Figure 9. About 10% of all the detec-

tions are false positives. The distribution of the met-

rics for the 20 cases (Figure 8) show no signiﬁcate

dispersion between quartiles, with only one outlier.

The results suggest the colours of interest are cor-

rectly identiﬁed.

The anatomical segmentation was applied on the

resulting background of the colour isolation of the

images. The evaluation is made against the manual

annotations of the left atrium. The mean values for

the performance metrics are: sensitivity of 0.805 and

false detection rate of 0.284.

More than 80% of the pixels belonging to left atria

were correctly identiﬁed as positives, which means

the framework achieved the purpose of deﬁning a

rough region of interest. However, the remaining ex-

cluded pixels may contain valuable colour informa-

tion for the regurgitation jet search. 28% of false pos-

itives indicate that some parts of the boundary were

Figure 8: Whiskers plot of sensitivity and false detection

rate for colour pattern detection.

Figure 9: Result of segmentation by the colour pattern seg-

mentation pipeline (left) and manual annotation of the jet

(right).

extended to neighbour chambers, which may increase

the false detections of regurgitation jets candidates.

Examples of both false positives and false negatives

can be seen in Figure 11.

The fact that the boundaries are often not repre-

sented on the ultrasound image is probably the main

source of error for this framework.

The distribution of the results (Figure 10) shows

some variability between cases, which may indicate

that acquisition conditions affect the performance of

the framework.

4.2 Regurgitation Jet Detection

The ﬁnal tests consist in applying the left atrium

masks obtained by the localizing active contours

framework as the spatial restriction for the colour

pattern segmentation, in order to understand the im-

pact caused on the pattern segmentation by the semi-

automatic atrium segmentation. The mean values for

the performance metrics are: sensitivity of 0.826 and

false detection rate of 0.228. The distribution of the

results (Figure 12) is similar to the distribution for the

left atrium segmentation. It is possible to infer that

the faults on the anatomical segmentation decrease

the true jet candidates’ detection and increase false

detections. Both cases can be seen on the examples in

Figure 13.

Between the jet candidates, only a small part is

identiﬁed with conﬁdence as mitral regurgitation. The

results for the detection of those candidates is: sensi-

BIOIMAGING 2017 - 4th International Conference on Bioimaging

Figure 10: Whiskers plot of sensitivity and false detection

rate for left atrium segmentation.

Figure 11: Results of anatomical segmentation (top) and

manual annotations (bottom). First column shows a satis-

factory result. Second column is an example of overesti-

mation of the atrium area. Third column is an example of

underestimation of atrium area.

tivity of 0.863 and false detection rate of 0.513. The

sensitivity increases when compared to the overall de-

tection of patterns. However, the false detection rate

is signiﬁcate, since the images contain a substantial

amount of non-regurgitation colour patterns.

Another possibility for false detection suppres-

sion, is to conﬁne the search for regurgitation to the

systolic phase of the cardiac cycle. The dataset was

manually identiﬁed as systolic or diastolic and the

pipeline was applied only on the systolic group. The

results for detection of real jets are: sensitivity of

0.840 and false detection rate of 0.355. As expected,

the false detection rate decreases if the diastolic phase

is not included, since all the candidates present on that

phase are certainly not regurgitation jets.

5 CONCLUSION

A new method for the segmentation of the mitral re-

gurgitation jet in sequences of 2D+t Doppler echocar-

diography was proposed. While the colour patterns of

interest are correctly identiﬁed as regurgitation candi-

dates, the anatomical spatial restriction step is still far

from ideal. Localizing region-based active contours

Figure 12: Whiskers plot of sensitivity and false detection

rate for the combined pipeline.

Figure 13: Results of combined pipeline (top) and manual

annotations (bottom). Occurrence of false negatives (left)

and false positives (right).

adapt to the heterogeneity of the left atrium, but do

not recover well from discontinuities in the bound-

ary of the atrium. This usually occurs on the part of

the boundary formed by the mitral valve, because of

the opening of the leaﬂets. The results conﬁrm that

the false positives can be decreased by limiting the

search to the systolic phase. The conﬁnement of the

search for that part of the cycle may also improve

the performance of the localizing region-based active

contour, since the leaﬂets are closed, forming the re-

quired boundary. However, the contour initialization

for each frame is currently the resulting mask from

the previous frame. The division of the sequence into

cycles will create the need for a re-initialization strat-

egy for each systole in a sequence

The addition of manual annotations by other sub-

jects may provide more certainty for the classiﬁcation

of the colour pattern candidates, and therefore reduce

the false positives. Regarding the colour, it is difﬁ-

cult to determine further characteristics for the dis-

tinction of candidates between regurgitation jet and

turbulence or other phenomena. Other characteristics

such as shape and relative position or orientation of

the candidates may be useful to decrease the false de-

tection ratio of the jet detection.

Future developments include the detection of the

A Preliminary Approach for the Segmentation of Mitral Valve Regurgitation Jet in Doppler Ecocardiography Images

cardiac cycle phase and inclusion of characteristics

other than colour for jet detection.

ACKNOWLEDGEMENTS

This article is a result of the project (NORTE-

01-0247-FEDER-003507-RHDecho), co-funded by

Norte Portugal Regional Operational Programme

(NORTE 2020), under the PORTUGAL 2020 Part-

nership Agreement, through the European Regional

Development Fund (ERDF). This work also had the

collaboration of the Fundac¸

ao para a Ci

encia a e Tec-

nologia (FCT) grant no: PD/BD/105761/2014 and has

contributions from the project NanoSTIMA, NORTE-

01-0145-FEDER-000016, supported by Norte Portu-

gal Regional Operational Programme (NORTE 2020),

through Portugal 2020 and the European Regional

Development Fund (ERDF). GE and PHILLIPS for

providing the equipment. Health professionals from

ırculo do Corac¸

ao for their volunteer work and data

collection. The Health Secretary of Para

ıba for their

support to the actualization of the Heart Caravan.

REFERENCES

B. Remenyi, N. Wilson, A. S. B. F.-J. K. K. K. J. L. G. M. E.

M. M. M. A. M. C. M. J. P. A. S. J. S. J. S. S. V. V. B.

G. W. L. Z. J. C. (2012). World heart federation crite-

ria for echocardiographic diagnosis of rheumatic heart

disease - an evidence-based guideline. In Nature Re-

views, Cardiology. I. Macmillan Publishers Limited.

D. Engelman, J. Kado, B. R. S. C.-J. C. N. W. S. D. A. S.

(2016). Screening for rheumatic heart disease: qual-

ity and agreement of focused cardiac ultrasound by

brieﬂy trained health workers. In BMC Cardiovascu-

lar Disorders. Springer Nature.

E. Marijon, D. Celermajer, M. T. S. E.-H. D. J. B. F.

X. J. (2009). Rheumatic heart disease screening by

echocardiography: The inadequacy of world health

organization criteria for optimizing the diagnosis of

subclinical disease. In Circulation. American Heart

Association.

J. Lu, C. Sable, G. E. C. W.-J. S. T. A. P. L. J. G. A. B.

(2015). Handheld echocardiographic screening for

rheumatic heart disease by non-experts. In Heart.

BMJ Publishing and British Cardiology Society.

M. Essop, F. P. (2014). Contemporary issues in rheumatic

fever and chronic rheumatic heart disease. In Circula-

tion. American Heart Association.

M. Gewitz, R. Baltimore, L. T. C. S.-S. S. J. C. B. R. K. T.

A. B. L. B. B. M. A. B. N. P. E. K. (2015). Revision of

the jones criteria for the diagnosis of acute rheumatic

fever in the era of doppler echocardiography. In Cir-

culation. American Heart Association.

M. Mirabel, R. Bacquelin, M. T. C. R.-B. H. P. C. I. F. K. N.

J. M. B. N. A. H. B. R. X. J. E. M. (2016). Screening

for rheumatic heart disease - evaluation of a focused

cardiac ultrasound approach. In Circ Cardiovascular

Imaging. American Heart Association.

M. Ploutz, J. Lu, J. S. C. W.-G. E. T. A. P. L. C. S. A. B.

(2015). Simpliﬁed rheumatic heart disease screening

criteria for handheld echocardiography. In Journal of

the American Society of Echocardiography. American

Society of Echocardiography.

R. Castello, P. Lenzen, A. L. (1992). Quantitation of mi-

tral regurgitation by transesophageal echocardiogra-

phy with doppler color ﬂow mapping: Correlation

with cardiac catheterization. In Journal of the Ameri-

can College of Cardiology. American College of Car-

diology.

R. Gonzalez, R. W. (2008). Digital Image Processing. Pear-

son Education, New Jersey, 3rd edition.

Roelandt, J. (1986). Color Doppler Flow Imaging and other

advances in Doppler echocardiography. Springer,

Netherlands, 1st edition.

S. Lankton, A. T. (2008). Localizing region- based active

contours. In IEEE Transactions on Image Processing.

S. Mazaheri, R. Wirza, P. S. M. D.-F. K. R. T. (2015). Seg-

mentation methods of echocardiography images for

left ventricle boundary detection. In Journal of Com-

puter Science.

BIOIMAGING 2017 - 4th International Conference on Bioimaging