Research Article Computer Vision Based Automatic...

Research ArticleComputer Vision Based Automatic Extraction and ThicknessMeasurement of Deep Cervical Flexor from Ultrasonic Images

Kwang Baek Kim1 Doo Heon Song2 and Hyun Jun Park3

1Department of Computer Engineering Silla University Busan 617-736 Republic of Korea2Department of Computer Games Yong-in Songdam College Yongin 449-040 Republic of Korea3Department of Computer Engineering Pusan National University Busan 609-735 Republic of Korea

Correspondence should be addressed to Kwang Baek Kim gbkimsillaackr

Received 5 July 2015 Revised 21 December 2015 Accepted 28 December 2015

Academic Editor Syoji Kobashi

Copyright copy 2016 Kwang Baek Kim et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

Deep Cervical Flexor (DCF) muscles are important in monitoring and controlling neck pain While ultrasonographic analysisis useful in this area it has intrinsic subjectivity problem In this paper we propose automatic DCF extractoranalyzer softwarebased on computer vision One of the major difficulties in developing such an automatic analyzer is to detect important organsand their boundaries under very low brightness contrast environment Our fuzzy sigma binarization process is one of the answersfor that problem Another difficulty is to compensate information loss that happened during such image processing proceduresMany morphologically motivated image processing algorithms are applied for that purpose The proposed method is verified assuccessful in extracting DCFs and measuring thicknesses in experiment using two hundred 800 times 600 DICOM ultrasonographyimages with 985 extraction rate Also the thickness of DCFs automatically measured by this software has small difference (lessthan 03 cm) for 898 of extracted DCFs

1 Introduction

Neck pain is very common complaint affecting up to 70 ofindividuals at some point of their lives [1] Once an individualdevelops neck pain it is reported that there is a 1 in 3chance that he or she will develop chronic symptoms lastinglonger than 6 months and the incidence of mechanical neckdisorders appears to be increasing [2] Clinical neck painis associated with impairment of muscle performance andthe functional impairments associated with neck pain andthe cause-effect relationships between neck pain and motorcontrol are well investigated [3]

The Deep Cervical Flexor (DCF) muscles includinglongus colli muscle longus capitis rectus capitis inferiorand rectus capitis lateralis have major roles in maintainingcervical lordosis and providing cervical joint stabilization[3 4] It has been theorized that when muscle performance isimpaired the balance between the stabilizers on the posterioraspect of the neck and the DCFs will be disrupted resulting

in loss of proper alignment and posture which is then likelyto contribute to cervical impairment [5]

A strong linear relation between the electromyographicamplitude of the DCF muscles and the incremental stagesof the craniocervical flexion test for control and individualswith neck pain was reported [6] Another study showed thatpatients with neck pain disorders have an altered neuromotorcontrol strategy during craniocervical flexion characterizedby reduced activity in the DCFs and increased activity inthe superficial flexors usually accompanied by altered move-ment strategies Furthermore they display reduced isometricendurance of the DCFs [7]

These observations prompted the use of the cranio-cervical flexion action for retraining the DCF muscles forneck pain patients Specific training of the DCF muscles inwomenwith chronic neck pain reduces pain and improves theactivation of these muscles especially in those with the leastactivation of their Deep Cervical Flexors before training [8]The recruitment pattern of the DCF and sternocleidomastoid

Hindawi Publishing CorporationComputational and Mathematical Methods in MedicineVolume 2016 Article ID 5892051 11 pageshttpdxdoiorg10115520165892051

2 Computational and Mathematical Methods in Medicine

is investigated during the craniocervical flexion test usingultrasonography [9 10] and it is further developed into anexercise program with tools [11]

Using ultrasound image in muscle analysis is appropri-ate for its noninvasive inexpensive real time respondingcapabilities [12] However its limitations are often pointedout sonographic images are dependent on the qualities ofequipment and skills of expertise thus the diagnosis oftenmisleads to subjective judgments [13] Thus we need anautomatic image segmentation and identification tool foranatomical landmarks that can eliminate such subjectivity inthe image analysis [14]

Unfortunately there is almost no directly related researchfor such an automatic DCF extractoranalyzer by computervision yet A recent study tried to give an automatic segmen-tation of cervical vertebrae from X-rays [15] but not relatedto muscles of our interests Probably our previous researchthat automatically detects sternocleidomastoid and longuscapitiscolli is the only one to consider [16]

In this paper we extend our previous automatic cervicalmuscle extractoranalyzer to accommodate withDCFs whichare longus capitis (Lcap) and longus colli (Lcol) Whilecertain part of extracting techniques is common to theprevious study that extracts sternocleidomastoid and longuscapitiscolli the overall treatment of target muscle DCF isdifferent from that of other related muscles due to the musclecharacteristics and the location of the muscles

The main methodological difference between our previ-ous study [16] and this work is whether the system extractsthe cervical vertebrae directly or not The brightness contrastbetweenDCF and related fascia ismuchweaker than betweenSCM and its related fascia in ultrasonography Also thecervical vertebrae located directly below the DCF havesimilar brightness values to those of DCFs Thus when weinfer the location of cervical vertebrae using the slope of lowerboundary lines of DCFs in [16] the key point to measurethe thickness of DCF is not correctly extracted especiallywhen the contrast between the fasciabones and DCFs is lowThus in this paper we extract cervical vertebrae directly andthe key points are induced based on the location of cervicalvertebrae

2 Overall Procedure

All the digital images used in this study are acquired andstored in DICOM (digital imaging and communications inmedicine) standard format DICOM is a form to declareimage transfer structure and related information In regionof interest (ROI) part of the image shown in Figure 1 there isa blood vessel and two muscles are located above and belowthe vessel The muscle above blood vessel is the sternocleido-mastoid and the muscle below the blood vessel is the DCFsDCFs have irregular form of border with cervical spine atthe bottom and is unclear being far from the center Thebrightness of themuscle is usually lower than that of the fasciaand the spine The fat area that has relatively high brightnessin the muscle area should be extracted with the muscle

Figure 2 summarizes the overall vision based system toextract and measure the thickness of DCFs automatically

Sternocleidomastoid muscle

Blood vessel

Deep neck flexor

Figure 1 ROI of ultrasound image

3 Extracting Deep Cervical Flexor

From this ROI image that contains only muscles fasciae andspines we try to extract candidate DCF by applying a seriesof image processing algorithms as shown in Figure 3

The ends-in search stretching method is a normalizationprocess that enhances the intensity contrast to differentiatetwo areas more clearly as shown in Figure 3(a) Formula (1)explains ends-in search stretching

119875 (119909 119910)

=

0 119866 (119909 119910) le Min

255 times

119866 (119909 119910) minusMinMax minusMin

Min lt 119866 (119909 119910) lt Max

255 119866 (119909 119910) ge Max

(1)

where Min and Max denote the maximum and minimumintensity value of the given image and 119866(119909 119910) is the intensityvalue of the pixel at (119909 119910) coordinates in the original imageand119875(119909 119910) is the resultant intensity value after ends-in searchstretching

The rationale of using ends-in search stretching is asfollows

Due to the scattering effect of ultrasound technologythere might be blurring that gives difficulty in discriminatingfascia muscles and cervical vertebrae Thus we need acontrast enhancing mechanism with minimal informationloss

In our method of extracting DCF we first extract thefascia lines and cervical vertebrae that are relatively brighterthan others and SCM and DCF are extracted by usinglocation characteristics related to the blood vessel A typicalcraniocervical sonography has largely skewed distribution ofintensity toward zero as shown in Figure 4 and those lowintensity pixels have slim chance to be related to objects of ourconcern Ends-in search stretching a type of normalizationprocess is then advantageous than other methods suchas filtering since it preserves the relative characteristics ofthe intensity distribution However it needs proper noiseremoval process afterwards

Computational and Mathematical Methods in Medicine 3

pointsExtracting key

thickness of DCFMeasuring

objectExtracting DCF

Remove fat area

binarizationFuzzy sigma

cervical vertebraecandidateExtracting

and boundariescandidate DCF

ExtractingPreprocessing

Figure 2 Overall system diagram

(a) Ends-in search stretching (b) Average binarization

(c) Removing noises and unnecessary holes

Figure 3 Extracting Deep Cervical Flexor candidate

0 255

Figure 4 Cervical vertebrae sonography and histogram


(a) Search for the object (b) Extracting upperlower boundary lines

(c) Filling holes to 1

Figure 5 Filling holes by searching boundaries of DCF

Next we apply average binarization to the enhancedimage shown in Figure 3(b) Then we apply Blob algo-rithm [17] to group pixels into objects Figure 3(b) containsunnecessary noise objects As explained in Section 5 indetail we obtained 200 images from 100 healthy subjectsduring a neck muscle endurance test We know that DCFsin consideration have specific morphological characteristicsthat we can use in noise removal process In order to quantifysuch characteristics that can be generalized we choose 50random samples from 200 image population and observe anysize or location related features that are common to apparentnoise objects in this situation

As shown in Figure 1 there exists blood vessel in betweenSCM and DCF thus two muscles are apart from a certaindistance From our samples the distance between twomuscleobjects in 119910-axis is no less than 85 of the ROI height Alsowe know that the lower fascia boundary lines and bones havea long curved shape thus it is not at the skewed location tothe left in any of our obtained images Thus we can set upa safe 119909-axis constraint such that an object that is no longerthan the half of the ROI width starts with the very leftmostposition (10 of the ROI width) Formula (2) below showsour noise removal criteria used in this paper

if ((119878 (119874) lt 0085 times 119878 (ROI)) OR(119874lef t lt 01 timesWidth (ROI) ANDWidth (119874) lt 05 timesWidth (ROI)))

then Remove (119874)

(2)

where 119878(119874) 119878(ROI) denote the size of the object in consider-ation and that of ROI area respectively andOleft denote the119909-coordinate of the leftmost boundary of the object 119874 andWidth(119883) is the function of returning the width of object119883

There might be some unnecessary holes in between theupper and lower boundary lines of candidate DCF In order

to fill such holes systematically we search upper and lowerboundary object by searching from the top and the bottom ofthe ROI area as shown in Figure 5 That results in an imagelike Figure 3(c)

Since we apply many image processing algorithms thatenhance the contrast of the brightness for purposes theremight exist cases where the boundary lines are disconnectedin part The first simple treatment for restoration is to fillbrightness values of pixels with 255 if they are neighbors of255-valued ones and makes the line connected The lowerboundary lines have relatively complex shape thus we apply4-directional contour search [17] to determine the boundarylines However again it may have discontinued part

Thus we apply cubic spline interpolation to reconnectlower boundaries The necessary conditions of cubic splineinterpolations are as follows

11987810158401015840

119899minus1(119875119899) = 11987810158401015840

0(1198750) = 0

11987810158401015840

119894minus1(119875119894) = 11987810158401015840

119894(119875119894) 119894 = 1 2 119899 minus 1

1198781015840

119894minus1(119875119894) = 1198781015840

119894(119875119894) 119894 = 1 2 119899 minus 1

119878119894minus1(119875119894) = 119878119894(119875119894) = 119910119894119894 = 1 2 119899 minus 1

(3)

where 119878119894denotes a Spline function and 119894 is the coordination

on the boundary and 119899 denotes the number of coordinationon the boundary and 119910

119894is the functional result of 119875

119894 Then

the Spline function 119878 is defined as follows

119878119894(119875) = 119886

119894+ 119887119894(119875 minus 119875

119894) + 119888119894(119875 minus 119875

119894)2

+ 119889119894(119875 minus 119875

119894)3

(4)

where 119886 119887 119888 and 119889 are constants satisfying Spline functionconditions [18]

Figure 6 demonstrates the extracted reconnected lowerboundaries of DCF Figure 6(a) and the overall shape of theextracted candidate of cervical vertebrae Figure 6(b)


(a) Lower boundary of DCF candidate (b) Candidate cervical vertebrae

Figure 6 Candidate cervical vertebrae extraction with image restoration

The next step is the binarization procedure The mainpurpose of binarization is to find an optimal thresholdingvalue to discriminate the target organ from background ofthe input image Our goal is to find the upper bound andlower bound lines of DCF area Unfortunately the brightnesscontrast near DCF area is not clear in general Thus we needmore detailed computationally heavy binarization procedurecalled fuzzy sigma binarization [19]

The major reason that we adopt computationally heavyfuzzy binarization over simple average binarization or Otsubinarization [20] is the environmental characteristic of thenear DCF area Since cervical vertebrae DCFs and relatedfascia have similar brightness values the average binarizationmay include extracting false positive objects in such a lowcontrasted environment Otsu binarization that assumesthe image contains two classes of pixels following bimodalhistogramand searches for the optimum threshold separatingthe two classes so that the interclass variance is minimal isanother alternative but in this environment the cervicalvertebrae are relatively brighter than nearby muscles andfascia thus the interclass variance should not be minimalsince we want to extract cervical vertebrae in this case Thusfuzzy sigma binarization that is designed to adapt the sen-sitivity of environment (brightness contrast) and qualitativemembership decision by fuzzy membership function is ourchoice Figure 7 demonstrates the effects of using differentbinarization methods in this environment

The fuzzy membership function of our fuzzy sigmabinarization is defined as follows

Let 119875Max 119875Min and 119875Mid be the highest and lowestbrightness value of the pixel and the average of 119875Max and119875Min respectively Then the membership function is definedas follows

Step 1 Consider the following

119875119865

Min = 119875Mid minus 119875Min

119875119865

Max = 119875Max minus 119875Mid(5)


if (119875Mid gt 119875Min + 075 (119875Max minus 119875Min)) then 119875119865

Mid

= 255 minus 119875Mid

else 119875119865Mid = 119875Mid

(6)


if (119875119865Mid gt 119875119865

Max) then

if (119875119865Min gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Min

else if (119875119865Max gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Max

(7)

Step 4 Calculate the normalized 119875newMin and 119875newMax

119875NewMin = 119875Mid minus 120573

119875NewMax = 119875Mid + 120573

(8)

Step 2 is designed to compensate overly brightly filmedimage due to filming environment while classifying 119875119865Midbased on the three quarters of brightness contrast

The membership degree of a pixel by sigma fuzzy bina-rization is as follows within the interval [119875NewMin 119875

NewMax ]

if (119875 le 119875NewMin ) then 119906 (119875) = 0

if (119875NewMin lt 119875 lt 119875NewMid ) then 119906 (119875)

=

119875 minus 119875NewMin

119875NewMid minus 119875

NewMin

if (119875NewMid ge 119875) then 119906 (119875) = 1

(9)

The membership degree 119906(119875) is applied to the a-cut (setto 05 in this paper since there is no prior information or


(a) Candidate cervical vertebrae (b) After fuzzy binarization

(c) After average binarization (d) After Otsu binarization

Figure 7 Different binarizations in candidate cervical vertebrae extraction

u(P) = 1

120572-cut

IPmin I

Pmid I

Pmax

Figure 8 Membership function for sigma fuzzy binarization

preferences) for the binarization such that the pixel valuewould be set to 255 if 119906(119875) ge 05 and 0 otherwise Figure 8shows the membership function of sigma type

Again we apply Blob algorithm to remove unnecessarynoises and apply expansion operation to restore small discon-tinuity if exists The result is shown in Figure 9

After expansion operation we can extract the lowerboundaries of DCF using relative location informationamong objects appearing in the cervical vertebrae areaSubcutaneous fat objects existing above DCFwill be removedfor the simplification of further analysis Possible disconnec-tions caused by fat removal are restored by simple DigitalDifferential Analyzer (DDA) algorithm [19] that is a type ofsimple linear interpolation of slope119898 between given intervalsusing the following equation

119876 (119909 119910) = 119898 (119909 minus 1199091) + 1199101

119898 =

1199102minus 1199101

1199092minus 1199091

(10)

where 119876(119909 119910) denotes the coordinates of a pixel in betweenthe left side of the extracted object and the right cervicalvertebrae The effect of DDA algorithm can be shown inFigure 10

The upper bound lines of Figure 10 are the lower bound-ary lines of DCF area Since all other objects such as subcu-taneous fat are removed already it is simple to find upperbound lines of Figure 10 However there might be small


(a) Fuzzy binarization applied (b) Noise removed

Figure 9 DCF area treatment for connectivity

Figure 10 Boundary reconnection of cervical vertebrae by DDAalgorithm

disconnections Again cubic spline interpolation defined as(3) and (4) is applied to compensate lost information

Figure 11 demonstrates the final extraction of lowerboundaries of DCF The final extraction result of DCF isshown in Figure 12

4 Measuring Thickness of DeepCervical Flexor

Among many possible characteristic features of DCF thethickness is the most fundamental one but it is also the mainsource of subjectivity in measuring morphometric featuresComputer vision based approach like us aims to locatemeasuring key points accurately to achieve the automaticityof capturing such measuring standards to avoid subjectivityas much as possible In our proposed method there are threecervical vertebrae in the image and themeasuring key point isset to be the rightmost point of the leftmost cervical vertebraeobjectThen from that key point two othermeasuring pointsare chosen to be 1 cm left and right to the key point Thethickness is then computed as the average of those threevertically measured lengths passing through the DCF Thedetails of such measuring process are as follows

The leftmost cervical vertebra should be located at theleft part of the image However when we apply the labeling

Figure 11 Extracting lower boundaries of DCF

Figure 12 Final extraction of DCF area

procedure to obtain target objects if the leftmost cervicalvertebra is found at the right part of the given image thatmeans the real leftmost cervical vertebra is lost as noise dur-ing preprocessing Then we apply weighted image restora-tion process to the image after fuzzy sigma binarization(Figure 9(a)) as shown in formula (11) to avoid unwantedobject removal

120572 = 10 +

radic(119909 minus 1199091)2

+ (119910 minus 1199101)2

1000

119875 (119909 119910) = 119875 (119909 119910) times 120572

(11)


(a) Candidate cervical vertebra (b) Lowest boundary of DCF

(c) After removing fat false positives

Figure 13 Fat area noise removal

Figure 14 Key point extraction

where 120572 is the weight and this formula gives more weight tothe points of the far left points from the center point (119909

1 1199101)

This compensation process is designed to avoid candidatecervical vertebra object located at the left side of the image

After applying Blob algorithm and object labeling pro-cedure if a candidate cervical vertebra is located above thelowest line of DCF it is probably the fat area falsely regardedas cervical vertebra thus we remove such false positives asshown in Figure 13

After such compensation process the measuring keypoint is set to be the rightmost point of the leftmost cervicalvertebrae object as shown in Figure 14

5 Experiment

The proposed method is implemented with C++ underMicrosoftVisual Studio 2010 on the IBM-compatible PCwithIntel Core i7-2600 CPU 340GHz and 4GB RAM Theexperiment uses two hundred (200) 800 times 600 size DICOMformat ultrasound images One hundred healthy subjects ofage in their 20s and 30s participated in this experiment

We measured neck muscle size during a neck muscleendurance test The thickness of neck muscle was assessedusing ultrasonography (MyLab 25GOLD)with 12MHz linearprobe during neck muscle endurance tests Sternocleidomas-toideus (SCM) and the deep cervical muscles which arelongus capitis (Lcap) and longus colli (Lcol) were measuredlaterally (right and left side) at the level of C4 Subjectsperformed a craniocervical flexion test (CCFT) The CCFTis commonly used in the physiotherapeutic assessment ofneck muscle strength and endurance During the CCFTsubjects were instructed to perform a nodding movementrepresenting the craniocervical flexion in five incrementallevels from 22 to 30mmHg 22 24 26 28 and 30mmHgThe subjects were asked to maintain the test positions for aslong as possible Each subject was asked to sit upright in anexamination chair with their arms resting on their thighs Ahead was fixed to maintain the head and neck in the neutralposition The C4 segmental level was chosen to optimize theimage field forCCFT and thus infer the coordination between


(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed

Figure 15 Comparison of muscle thickness measurements (cm)

the threemuscles in themidcervical spine At the C4 level wecaptured different parts of the Lcol Lcap and SCM

Figure 15 visually demonstrates the result of the pro-posed automatic DCF extractions (c1 and c2) compared withphysical therapistrsquos manual inspections (a1 and a2) and ourprevious attempts (b1 and b2) for the same images Onecan verify that our vision based automatic software extractsalmost identical key pointsarea of target muscles and themagnitude of difference between automatic measurementandmanualmeasurement becomesmuch smaller than that ofprevious attempt In our experiment twomedical experts areinvolved and the thickness of the manual inspection is takenas the average of the two human expertsrsquo evaluations

As summarized in Table 1 the proposed method is muchbetter than the previous attempt [16] in DCF extractionin sensitivity (SS) where TP TN FP and FN denote the

true positive true negative false positive and false negativerespectively

Due to the image acquisition process explained abovethere is no image that does not contain DCF and SCMThusthere is no TN or FN The decision to classify the result ofautomatic extractions is based on the agreement of two fieldexperts involved in this experiment

Table 2 also shows the improved sensitivity in SCMextractions

Also after measuring the thickness we compare ourresult with the previous attempt as the absolute differencefrom human expertsrsquo measurement (error) The results forthe error magnitude of thickness measurement for DCF andSCM are summarized in Tables 3 and 4 respectively

While much improved compared to the previous attempt[16] the distribution of error magnitude in DCF thickness


Table 1 DCF extraction performance evaluation

Success rate DCF (longus capitiscolli)TP TN FP FN SS

Previous [16] 147 0 53 0 735Proposed 197 0 3 0 985

Table 2 SCM extraction performance evaluation

Success rate SCMTP TN FP FN SS


Table 3 Error magnitude of DCF (longus capitiscolli) thickness (of successes of tried images)

lt01 cm lt02 cm lt03 cm ge03 cm

Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)

Table 4 Error magnitude of SCM thickness ( of successes oftried images)

Error lt 01 cm Error ge 01 cm

Previous [16] 141169(834)

28169(166)

Proposed 175198(884)

23198(116)

measurement shows that there is a room for the improvementin the future At least the proposed method is successful inextracting DCF and SCM accurately and the sensitivity ofsuch extractions is much more improved compared to theprevious attempt

6 Conclusion

In this paper we propose a vision based fully automaticmethod to extract DCF muscles (longus capitiscolli) andSCM with measuring the thickness from cervical verte-brae ultrasound images DCF muscles are important incontrollingmonitoring the neck pain andor develop effi-cienteffective rehabilitative training procedures Howeverexisting clinical method using ultrasonography often causesoperator effect a subjective judgments of muscle extractionand associated morphometric features measurement andanalysis Our attempt shown in this paper is to avoid suchsubjectivity as much as possible and aims to assist humanmedical experts in DCF and SCM analysis

Algorithmically our previous attempt [16] that extractsand analyzes sternocleidomastoid and longus capitiscolliautomatically often suffers from the distraction of ultra-sonography The major contribution of this paper is to over-come that problemwith careful image processing considering

morphological characteristics of DCFs (shape and locationinformation) and extracting cervical vertebrae explicitly inthe process Many image processing algorithms are involvedsuch as Blob for noise removal and ends-in search stretchingfor image enhancement and 4-directional contour searchfor boundary detection and cubic spline interpolation forcomplementing disconnected lines and the fuzzy sigma bina-rization to control the low brightness contrast environment

The experimental result using 200 real-world cervicalvertebrae DICOM ultrasound image verifies that almost all(985) input images are well extracted and the thicknessis automatically measured with low error magnitude withrespect to those of human expertsrsquo decision (le03 cm) inmostcases (898)

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

References

[1] R Fejer K O Kyvik and J Hartvigsen ldquoThe prevalence of neckpain in the world population a systematic critical review of theliteraturerdquo European Spine Journal vol 15 no 6 pp 834ndash8482006

[2] ANygrenA Berglund andMVonKoch ldquoNeck-and-shoulderpain an increasing problem Strategies for using insurancematerial to follow trendsrdquo Scandinavian Journal of Rehabilita-tion Medicine Supplement vol 32 pp 107ndash112 1995

[3] D Falla and D Farina ldquoNeural and muscular factors associatedwith motor impairment in neck painrdquo Current RheumatologyReports vol 9 no 6 pp 497ndash502 2007

[4] L C Boyd-Clark C A Briggs and M P Galea ldquoMusclespindle distribution morphology and density in longus colliand multifidus muscles of the cervical spinerdquo Spine vol 27 no7 pp 694ndash701 2002

[5] Z A Iqbal R Rajan S A Khan and A H Alghadir ldquoEffectof deep cervical flexor muscles training using pressure biofeed-back on pain and disability of school teachers with neck painrdquoJournal of Physical Therapy Science vol 25 no 6 pp 657ndash6612013

[6] D L Falla G A Jull and P W Hodges ldquoPatients withneck pain demonstrate reduced electromyographic activity ofthe deep cervical flexor muscles during performance of thecraniocervical flexion testrdquo Spine vol 29 no 19 pp 2108ndash21142004

[7] G A Jull S P OrsquoLeary and D L Falla ldquoClinical assessment ofthe deep cervical flexor muscles the craniocervical flexion testrdquoJournal of Manipulative and Physiological Therapeutics vol 31no 7 pp 525ndash533 2008

[8] D Falla S OrsquoLeary D Farina and G Jull ldquoThe change indeep cervical flexor activity after training is associated with thedegree of pain reduction in patients with chronic neck painrdquoThe Clinical Journal of Pain vol 28 no 7 pp 628ndash634 2012

[9] F M R Jesus P H Ferreira and M L Ferreira ldquoUltrasono-graphic measurement of neck muscle recruitment a prelimi-nary investigationrdquo Journal of Manual ampManipulativeTherapyvol 16 no 2 pp 89ndash92 2008

[10] I Jun and K Kim ldquoA comparison of the deep cervical flexormuscle thicknesses in subjects with and without neck pain


during craniocervical flexion exercisesrdquo Journal of PhysicalTherapy Science vol 25 no 11 pp 1373ndash1375 2013

[11] K H Yun and K Kim ldquoEffect of craniocervical flexion exerciseusing sling on thickness of sternocleidomastoid muscle anddeep cervical flexor musclerdquo Journal of the Korean Society ofPhysical Medicine vol 8 no 2 pp 253ndash261 2013

[12] E Cardinal N J Bureau B Aubin and R K Chhem ldquoRole ofultrasound in musculoskeletal infectionsrdquo Radiologic Clinics ofNorth America vol 39 no 2 pp 191ndash201 2001

[13] R J Mason V C Broaddus T Martin et alMurray and NadelrsquosTextbook of Respiratory Medicine 2-Volume Set Elsevier HealthSciences 2010

[14] E T Enikov and R Anton ldquoImage segmentation and analysisof flexion-extension radiographs of cervical spinesrdquo Journal ofMedical Engineering vol 2014 Article ID 976323 9 pages 2014

[15] X Xu H-W Hao X-C Yin N Liu and S H ShafinldquoAutomatic segmentation of cervical vertebrae in X-ray imagesrdquoin Proceedings of the Annual International Joint Conference onNeural Networks (IJCNN rsquo12) pp 1ndash8 IEEE Brisbane AustraliaJune 2012

[16] K B Kim H J Park D H Song and S-S Han ldquoExtractionof sternocleidomastoid and longus capitiscolli muscle usingcervical vertebrae ultrasound imagesrdquo Current Medical ImagingReviews vol 10 no 2 pp 95ndash104 2014

[17] M Petrou and P Bosdogianni Image Processing John Wiley ampSons 1999

[18] K B Kim H J Lee D H Song and Y W Woo ldquoExtractingfascia and analysis of muscles from ultrasound images withFCM-based quantization technologyrdquo Neural Network Worldvol 20 no 3 pp 405ndash416 2010

[19] K B Kim D H Song and W J Lee ldquoFlaw detection inceramics using sigma fuzzy binarization and gaussian filteringmethodrdquo International Journal of Multimedia and UbiquitousEngineering vol 9 no 1 pp 403ndash414 2014

[20] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979

Submit your manuscripts athttpwwwhindawicom

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinsonrsquos Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom


is investigated during the craniocervical flexion test usingultrasonography [9 10] and it is further developed into anexercise program with tools [11]

Using ultrasound image in muscle analysis is appropri-ate for its noninvasive inexpensive real time respondingcapabilities [12] However its limitations are often pointedout sonographic images are dependent on the qualities ofequipment and skills of expertise thus the diagnosis oftenmisleads to subjective judgments [13] Thus we need anautomatic image segmentation and identification tool foranatomical landmarks that can eliminate such subjectivity inthe image analysis [14]

Unfortunately there is almost no directly related researchfor such an automatic DCF extractoranalyzer by computervision yet A recent study tried to give an automatic segmen-tation of cervical vertebrae from X-rays [15] but not relatedto muscles of our interests Probably our previous researchthat automatically detects sternocleidomastoid and longuscapitiscolli is the only one to consider [16]

In this paper we extend our previous automatic cervicalmuscle extractoranalyzer to accommodate withDCFs whichare longus capitis (Lcap) and longus colli (Lcol) Whilecertain part of extracting techniques is common to theprevious study that extracts sternocleidomastoid and longuscapitiscolli the overall treatment of target muscle DCF isdifferent from that of other related muscles due to the musclecharacteristics and the location of the muscles

The main methodological difference between our previ-ous study [16] and this work is whether the system extractsthe cervical vertebrae directly or not The brightness contrastbetweenDCF and related fascia ismuchweaker than betweenSCM and its related fascia in ultrasonography Also thecervical vertebrae located directly below the DCF havesimilar brightness values to those of DCFs Thus when weinfer the location of cervical vertebrae using the slope of lowerboundary lines of DCFs in [16] the key point to measurethe thickness of DCF is not correctly extracted especiallywhen the contrast between the fasciabones and DCFs is lowThus in this paper we extract cervical vertebrae directly andthe key points are induced based on the location of cervicalvertebrae

2 Overall Procedure

All the digital images used in this study are acquired andstored in DICOM (digital imaging and communications inmedicine) standard format DICOM is a form to declareimage transfer structure and related information In regionof interest (ROI) part of the image shown in Figure 1 there isa blood vessel and two muscles are located above and belowthe vessel The muscle above blood vessel is the sternocleido-mastoid and the muscle below the blood vessel is the DCFsDCFs have irregular form of border with cervical spine atthe bottom and is unclear being far from the center Thebrightness of themuscle is usually lower than that of the fasciaand the spine The fat area that has relatively high brightnessin the muscle area should be extracted with the muscle

Figure 2 summarizes the overall vision based system toextract and measure the thickness of DCFs automatically

Sternocleidomastoid muscle

Blood vessel

Deep neck flexor

Figure 1 ROI of ultrasound image

3 Extracting Deep Cervical Flexor

From this ROI image that contains only muscles fasciae andspines we try to extract candidate DCF by applying a seriesof image processing algorithms as shown in Figure 3

The ends-in search stretching method is a normalizationprocess that enhances the intensity contrast to differentiatetwo areas more clearly as shown in Figure 3(a) Formula (1)explains ends-in search stretching

119875 (119909 119910)

=

0 119866 (119909 119910) le Min

255 times

119866 (119909 119910) minusMinMax minusMin

Min lt 119866 (119909 119910) lt Max

255 119866 (119909 119910) ge Max

(1)

where Min and Max denote the maximum and minimumintensity value of the given image and 119866(119909 119910) is the intensityvalue of the pixel at (119909 119910) coordinates in the original imageand119875(119909 119910) is the resultant intensity value after ends-in searchstretching

The rationale of using ends-in search stretching is asfollows

Due to the scattering effect of ultrasound technologythere might be blurring that gives difficulty in discriminatingfascia muscles and cervical vertebrae Thus we need acontrast enhancing mechanism with minimal informationloss

In our method of extracting DCF we first extract thefascia lines and cervical vertebrae that are relatively brighterthan others and SCM and DCF are extracted by usinglocation characteristics related to the blood vessel A typicalcraniocervical sonography has largely skewed distribution ofintensity toward zero as shown in Figure 4 and those lowintensity pixels have slim chance to be related to objects of ourconcern Ends-in search stretching a type of normalizationprocess is then advantageous than other methods suchas filtering since it preserves the relative characteristics ofthe intensity distribution However it needs proper noiseremoval process afterwards





Remove fat area









0 255










(2)






11987810158401015840

119899minus1(119875119899) = 11987810158401015840

0(1198750) = 0

11987810158401015840

119894minus1(119875119894) = 11987810158401015840

119894(119875119894) 119894 = 1 2 119899 minus 1

1198781015840

119894minus1(119875119894) = 1198781015840

119894(119875119894) 119894 = 1 2 119899 minus 1

119878119894minus1(119875119894) = 119878119894(119875119894) = 119910119894119894 = 1 2 119899 minus 1

(3)




119894 Then


119878119894(119875) = 119886

119894+ 119887119894(119875 minus 119875

119894) + 119888119894(119875 minus 119875

119894)2

+ 119889119894(119875 minus 119875

119894)3

(4)











119875119865


119875119865




Mid

= 255 minus 119875Mid

else 119875119865Mid = 119875Mid

(6)


if (119875119865Mid gt 119875119865

Max) then

if (119875119865Min gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Min

else if (119875119865Max gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Max

(7)


119875NewMin = 119875Mid minus 120573

119875NewMax = 119875Mid + 120573

(8)



NewMax ]

if (119875 le 119875NewMin ) then 119906 (119875) = 0


=



NewMin

if (119875NewMid ge 119875) then 119906 (119875) = 1

(9)






u(P) = 1

120572-cut

IPmin I

Pmid I

Pmax





119876 (119909 119910) = 119898 (119909 minus 1199091) + 1199101

119898 =

1199102minus 1199101

1199092minus 1199091

(10)















120572 = 10 +

radic(119909 minus 1199091)2

+ (119910 minus 1199101)2

1000

119875 (119909 119910) = 119875 (119909 119910) times 120572

(11)







1 1199101)




5 Experiment




(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




















Remove fat area









0 255










(2)






11987810158401015840

119899minus1(119875119899) = 11987810158401015840

0(1198750) = 0

11987810158401015840

119894minus1(119875119894) = 11987810158401015840

119894(119875119894) 119894 = 1 2 119899 minus 1

1198781015840

119894minus1(119875119894) = 1198781015840

119894(119875119894) 119894 = 1 2 119899 minus 1

119878119894minus1(119875119894) = 119878119894(119875119894) = 119910119894119894 = 1 2 119899 minus 1

(3)




119894 Then


119878119894(119875) = 119886

119894+ 119887119894(119875 minus 119875

119894) + 119888119894(119875 minus 119875

119894)2

+ 119889119894(119875 minus 119875

119894)3

(4)











119875119865


119875119865




Mid

= 255 minus 119875Mid

else 119875119865Mid = 119875Mid

(6)


if (119875119865Mid gt 119875119865

Max) then

if (119875119865Min gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Min

else if (119875119865Max gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Max

(7)


119875NewMin = 119875Mid minus 120573

119875NewMax = 119875Mid + 120573

(8)



NewMax ]

if (119875 le 119875NewMin ) then 119906 (119875) = 0


=



NewMin

if (119875NewMid ge 119875) then 119906 (119875) = 1

(9)






u(P) = 1

120572-cut

IPmin I

Pmid I

Pmax





119876 (119909 119910) = 119898 (119909 minus 1199091) + 1199101

119898 =

1199102minus 1199101

1199092minus 1199091

(10)















120572 = 10 +

radic(119909 minus 1199091)2

+ (119910 minus 1199101)2

1000

119875 (119909 119910) = 119875 (119909 119910) times 120572

(11)







1 1199101)




5 Experiment




(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
























(2)






11987810158401015840

119899minus1(119875119899) = 11987810158401015840

0(1198750) = 0

11987810158401015840

119894minus1(119875119894) = 11987810158401015840

119894(119875119894) 119894 = 1 2 119899 minus 1

1198781015840

119894minus1(119875119894) = 1198781015840

119894(119875119894) 119894 = 1 2 119899 minus 1

119878119894minus1(119875119894) = 119878119894(119875119894) = 119910119894119894 = 1 2 119899 minus 1

(3)




119894 Then


119878119894(119875) = 119886

119894+ 119887119894(119875 minus 119875

119894) + 119888119894(119875 minus 119875

119894)2

+ 119889119894(119875 minus 119875

119894)3

(4)











119875119865


119875119865




Mid

= 255 minus 119875Mid

else 119875119865Mid = 119875Mid

(6)


if (119875119865Mid gt 119875119865

Max) then

if (119875119865Min gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Min

else if (119875119865Max gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Max

(7)


119875NewMin = 119875Mid minus 120573

119875NewMax = 119875Mid + 120573

(8)



NewMax ]

if (119875 le 119875NewMin ) then 119906 (119875) = 0


=



NewMin

if (119875NewMid ge 119875) then 119906 (119875) = 1

(9)






u(P) = 1

120572-cut

IPmin I

Pmid I

Pmax





119876 (119909 119910) = 119898 (119909 minus 1199091) + 1199101

119898 =

1199102minus 1199101

1199092minus 1199091

(10)















120572 = 10 +

radic(119909 minus 1199091)2

+ (119910 minus 1199101)2

1000

119875 (119909 119910) = 119875 (119909 119910) times 120572

(11)







1 1199101)




5 Experiment




(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
























119875119865


119875119865




Mid

= 255 minus 119875Mid

else 119875119865Mid = 119875Mid

(6)


if (119875119865Mid gt 119875119865

Max) then

if (119875119865Min gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Min

else if (119875119865Max gt 119875119865

Mid) then 120573 = 119875119865

Mid

else 120573 = 119875119865Max

(7)


119875NewMin = 119875Mid minus 120573

119875NewMax = 119875Mid + 120573

(8)



NewMax ]

if (119875 le 119875NewMin ) then 119906 (119875) = 0


=



NewMin

if (119875NewMid ge 119875) then 119906 (119875) = 1

(9)






u(P) = 1

120572-cut

IPmin I

Pmid I

Pmax





119876 (119909 119910) = 119898 (119909 minus 1199091) + 1199101

119898 =

1199102minus 1199101

1199092minus 1199091

(10)















120572 = 10 +

radic(119909 minus 1199091)2

+ (119910 minus 1199101)2

1000

119875 (119909 119910) = 119875 (119909 119910) times 120572

(11)







1 1199101)




5 Experiment




(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




















u(P) = 1

120572-cut

IPmin I

Pmid I

Pmax





119876 (119909 119910) = 119898 (119909 minus 1199091) + 1199101

119898 =

1199102minus 1199101

1199092minus 1199091

(10)















120572 = 10 +

radic(119909 minus 1199091)2

+ (119910 minus 1199101)2

1000

119875 (119909 119910) = 119875 (119909 119910) times 120572

(11)







1 1199101)




5 Experiment




(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




























120572 = 10 +

radic(119909 minus 1199091)2

+ (119910 minus 1199101)2

1000

119875 (119909 119910) = 119875 (119909 119910) times 120572

(11)







1 1199101)




5 Experiment




(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of






















1 1199101)




5 Experiment




(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















(a1) SCM 051 DCF 151 (a2) SCM 065 DCF 166

(a) Manual

(b1) SCM 056 DCF 167 (b2) SCM 069 DCF 184

(b) Previous [16]

(c1) SCM 053 DCF 156 (c2) SCM 066 DCF 171

(c) Proposed



















Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

























Previous [16] 3147(20)

32147(218)

72147(490)

40147(272)

Proposed 13197(66)

59197(299)

105197(533)

20197(102)



Previous [16] 141169(834)

28169(166)


23198(116)


6 Conclusion







References




























of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Research Article Computer Vision Based Automatic...

Documents

Transcript of Research Article Computer Vision Based Automatic...