Math & Engineering

A semiautomated ultrasound border detection program that facilitates clinical measurement of ultrasound carotid intima-media thickness

Description
A semiautomated ultrasound border detection program that facilitates clinical measurement of ultrasound carotid intima-media thickness
Published
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
   A Semiautomated Ultrasound Border Detection Program That Facilitates ClinicalMeasurement of Ultrasound CarotidIntima–Media Thickness James H. Stein, MD, Claudia E. Korcarz, DVM, RDCS, Maureen E. Mays, MD, MS,Pamela S. Douglas, MD, Mari Palta, PhD, Hongling Zhang, MS, Tamara LeCaire, MS,Diane Paine, MED, RDCS, David Gustafson, PhD, and Liexiang Fan, PhD,  Madison,Wisconsin, and Issaquah, Washington   We have developed a novel, semiautomated carotid intima–media thickness (CIMT) border detection program (AUTO) and evaluated its measurement reproducibility and accuracy. Images from 6 carotid segments were acquired in 50 subjects, for a total of 300 segments. Mean and maximum CIMT values were measured blindly at a reference (REF) lab and in duplicate by experienced (EXP) and novice (NOV)readers using manual (MAN) and AUTO methods.Coefficients of variation for AUTO measurements of mean (3.2%) and maximum (4.1%) CIMT were low,and the AUTO method improved the NOV reader’sreproducibility. Compared with the REF lab, mean (0.012    0.006 mm) and maximum (0.144    0.006mm) CIMT biases were small and equivalent to thoseof the REF lab (  P  <  .001). The AUTO method short-ened reading times by 35% to 46% (  P  <  .001). Weconclude that our novel AUTO CIMT measurement program improved reproducibility and was accu-rate. Compared with MAN tracing, the AUTO method agreed better with the REF lab and decreased read-ing time.  (J Am Soc Echocardiogr 2005;18:244-51.) U ltrasound measurement of carotid intima–mediathickness (CIMT) is a noninvasive and highly repro-ducible tool for detecting and quantifying subclini-cal atherosclerosis. 1 Several large prospective epide-miologic and treatment studies have shown thatCIMT accurately identifies incident cardiovascular disease. 1-5 Despite recommendations by the Ameri-can Heart Association, however, CIMT testing is notused widely in clinical practice, because interpreta-tion of CIMT scans requires very precise measure-ments that can be time-consuming. 1 To help addressthese barriers to the clinical implementation of CIMT as a risk assessment tool, we developed anovel, semiautomated CIMT border detection pro-gram and evaluated the reproducibility of its mea-surements, its accuracy compared with that of areference laboratory, and its ability to decreasereading time compared with manual tracing. METHODS Experimental Protocol  The Institutional Review Board of the University of Wis-consin Medical School approved the design of this study.Data were obtained from 25 consecutive healthy individ-uals, mean age 28 years (standard deviation [SD], 8 years), who were imaged as part of the Cardiovascular Disease inType 1 Diabetes (Cardio-Diab) Study. An additional 25consecutive subjects (mean age, 57 years; SD, 8 years) who were referred by their physicians to the University of  Wisconsin Vascular Health Screening Program for deter-mination of CIMT were also recruited, to provide a wider range of CIMT values for this validation study. Carotid Ultrasound Imaging  The standardized protocol from the Atherosclerosis Risk in Communities study was used to acquire images of thefar walls of each carotid artery. 6 The common carotid From the Atherosclerosis Imaging Research Program, Section of Cardiovascular Medicine, and Department of Population HealthSciences, University of Wisconsin Medical School, Madison, Wis-consin and Siemens Medical Solutions, Issaquah, Washington.Supported by a grant from Siemens Medical Solutions. Dr. Stein was supported by the National Center for Research Resources(grant K23 RR16176), receives research support from SiemensMedical Solutions, and is a consultant to Camtronics MedicalSystems and Sonosite, Inc. Dr. Stein, Dr. Palta, Ms. Zhang, andMs.LeCairealsoreceivedfundingfromtheNationalHeart,Lung,and Blood Institute (grant R01 HL62897). Dr. Mays was sup-ported by National Institutes of Health National Service AwardT32 HL07936, from the University of Wisconsin-Madison Car-diovascular Research Center. Ms. Paine, Dr. Gustafson, and Dr.Fan are employees of Siemens Medical Solutions.Reprint requests: James H. Stein, MD, Department of Medicine,Section of Cardiovascular Medicine, University of WisconsinMedical School, 600 Highland Avenue, G7/341 CSC (MC3248), Madison, WI 53792 (E-mail:  jhs@medicine.wisc.edu  ).0894-7317/$30.00Copyright 2005 by the American Society of Echocardiography.doi:10.1016/j.echo.2004.12.002 244  artery (CCA) segment was defined as the distal 1 cm of theCCA, immediately proximal to the onset of increasedspatial separation of the walls of the CCA (ie, before thesrcin of the bulb). The carotid bifurcation segment wasdefined as the distal 1 cm of the bulb, the termination of  which was characterized by the presence of the flow divider between the internal carotid artery (ICA) and theexternal carotid artery (ECA). The ICA segment wasdefined as the proximal 1 cm of the ICA, starting imme-diately beyond the flow divider. Ultrasound images wereacquired using an 8.0-MHz linear array transducer (AcusonSequoia; Siemens Medical Solutions, Malvern, Penn) and were recorded digitally using a Camtronics Vericis acqui-sition module (Camtronics Medical Systems, Hartland, Wisc). CIMT Measurements The mean and maximum combined thicknesses of theintimal and medial layers of the far walls of each carotidartery segment were measured by an experienced reader (C.E.K.) and a novice (M.E.M.) reader, on 4 separateoccasions – twice using manual (MAN) tracing (CIMTScreen; Camtronics Medical Systems) and twice using aprototype semiautomated (AUTO) border detection pro-gram (Siemens Medical Solutions) ( Figure 1 ).  After images were acquired, the reader chose a single,end-diastolic still frame image for measurement of CIMT.Input for the AUTO program is the user’s single mouseclick in the vessel lumen at the left-sided limit of thearterial segment to be measured. The software automati-cally determines a region of interest (ROI) based on imageintensity, vessel morphology, and any user-defined seg-ment length. The ROI contains a portion of the lumen andthe corresponding intimal, medial, and adventitial layers.Next, the lumen–intima and media–adventitia borders within the ROI are automatically detected based on imageintensity and gradient information. The detected border points are processed to obtain smoothed borders withoutoutlier points. For each of the points along the media–adventitia border, the intima–media thickness is estimatedas the length of a line orthogonal to the media–adventitiaborder that ends at the lumen–intima border. The meanCIMT is the arithmetic mean value of these lengths. Themaximum CIMT is the maximum value of these lengths.The number of pixel points in a 1-cm-long segmentdepends on the image acquisition parameters. When8-MHz transmit frequency and 4-cm depth field of view areused, a 1-cm border contains 85 pixels; therefore, themean and maximum CIMT values would be estimatedfrom 85 data samples. If a region of the border is notacceptable, then a single click near the detected border enables rapid editing of the border. A total of 300 segmental and 100 composite CIMT values were compared between readers. The experiencedreader (EXP) previously had read more than 250 CIMTscans using the manual tracing method. The novice (NOV)had never read a CIMT scan before. All readings wereperformed with each reader blinded to previous measure-ments using either technique. Far-wall mean and maxi-mum CIMT were measured in triplicate for each carotidsegment; these measurements were averaged to deter-mine segmental values. The time to make all measure-ments, including necessary edits, was determined by each reader. In addition, the first 40 studies were sent on digitalmedia for independent blinded reading at a reference(REF) laboratory (Center for Medical Ultrasound CIMTReading Center, Wake Forest University School of Medi-cine, Winston-Salem, NC), resulting in 240 segmental and80 composite CIMT comparisons. 7 Composite mean andmaximum CIMT values were calculated as the respectiveaverages of the segmental mean and maximum valuesfrom all measurable segments (maximum of 6 per sub- ject). Statistical Techniques Mean (SD) values for segmental and composite mean andmaximum CIMT were calculated separately across allsubjects for readings 1 and 2 provided by both readers(EXP and NOV), using both systems (MAN and AUTO),and for measurements from the REF laboratory. For ease of reporting, composite mean and maximum CIMTs but only right CCA, bulb, and ICA CIMT values are given in thetables, because the left CIMT values were very similar.Important findings regarding measurements of left-sidedsegments are described in the text. Statistical assessments were performed using data from all segments. Figure 1  Image of monitor screen using the semiauto-mated border detection program. Measurement of far wallCIMT using the semiautomated border detection pro-gram. The superior traced line represents the blood–intimainterface; the bottom line represents the media–adventitiainterface identified by the semiautomated border detectionprogram. In this example, the maximum CIMT was .933mm, and the mean CIMT was .752 mm. Journal of the American Society of Echocardiography  Volume 18 Number 3  Stein et al  245  To evaluate intraobserver reproducibility, the mean(standard error of the mean [SEM]) arithmetic differencesin CIMT values (reading 2    reading 1) were analyzedusing the paired  t  -test. Pearson correlation coefficients with   P   values were also determined. The coefficient of  variation (CV) was calculated as the SD divided by themean using the root mean squared approach; 8 specifically,CV(  %  )     (   x  i1   x  i2  ) 2 2(   x  i1   x  i2  ) 2 2  n  1⁄2   100 where  x  i  1  is the first reading from the  i  th pair,  i    1,2,3,. . . n ,  x  i   2  is the second reading from the  i  th pair, and n  is the number of pairs used for comparisons.To compare CIMT measurements with the REF labora-tory, Pearson correlation coefficients with   P   values weredetermined, and values were displayed graphically usingthe method of Bland and Altman with the REF lab mea-surements on the  x  -axis. 9 The first set of measurementsfrom each reader using each technique was used for comparisons. The mean (SEM) arithmetic differences inCIMT values (reading 1    REF) were tested using theTOST (“two one-sided tests”) approach. 10,11 Predeter-mined levels of acceptable differences (    ) between each reader (EXP and NOV) and the REF laboratory were 0.11mm (1 digital pixel) for mean CIMT and 0.22 mm for maximum CIMT. The null hypothesis (H 0 , the means aredifferent) and alternate hypothesis (H 1  ) for the TOST aredescribed by the following formulas:  H  0  :   1   2    H  1  :   1   2   These formulae represent 2 separate null hypotheses. Thefirst null hypothesis is that the first mean (eg, of readingsby the EXP reader using the AUTO method) is less than thesecond mean (eg, of readings by the REF lab) by more thanthe acceptable amount,  . The second null hypothesis isthat the first mean is greater than the second mean by more than the acceptable amount,   . These hypotheses were tested by 2 separate 1-sided paired  t  -tests. If both hypotheses were rejected, then the hypothesis that clini-cally nonacceptable bias in readings is present is rejected. An  of 0.05 was used for all statistical tests. RESULTSSubject Characteristics The study group comprised 32 males and 18 fe-males, with a mean age of 42.7 (16.8) years. Averagesegmental and composite carotid intima–mediathickness measurements for the EXP reader usingthe AUTO system are presented in Table 1. These values represent the first reading by the EXP reader and are presented as reference for the comparisonsdetailed later. The composite mean CIMT was 0.707(0.170) mm. The composite maximum CIMT was0.957 (0.226) mm. Fourteen subjects had a total of 15 plaques (9 in the bulb and 6 in the internalcarotid artery). Intraobserver Reproducibility: Mean CIMT,Experienced Reader  Using the MAN method, composite mean CIMTreadings by the EXP reader correlated strongly (  r   .986,  P     .001) with a nonsignificant mean differ-ence of     0.001    0.004 mm and a CV of 3.1%,reflecting this reader’s extensive experience andfamiliarity with the MAN method software ( Table 2 ). Similar values were obtained for all 6 segmentalmean CIMT measurements. Using the AUTOmethod, composite mean CIMT readings by the EXPreader also correlated strongly (  r   .982,  P   .001) with a nonsignificant mean difference of 0.002   0.004 mm and a CV of 3.2%, indicating similar performance to the MAN method, despite unfamil-iarity with the AUTO method software. Similar val-ues were obtained for all 6 segmental mean CIMTmeasurements. Intraobserver Reproducibility: Maximum CIMT, Experienced Reader  Using the MAN method, composite maximum CIMTreadings correlated strongly (  r     .950,  P     .001) with a nonsignificant mean difference of 0.001   0.010 mm and a CV of 4.9%, also reflecting thisreader’s extensive experience with the MANmethod software ( Table 2 ). Similar, but slightly  higher, CVs were obtained for all 6 segmental max-imum CIMT measurements. Using the AUTOmethod, composite maximum CIMT readings alsocorrelated strongly (  r     .969,  P     .001) with anonsignificant mean difference of 0.005    0.008mm and a slightly better CV of 4.1%. Using the AUTO method yielded a notable improvement in theCV of right CCA maximum CIMT measurements. CV  values for the other 5 maximum segmental CIMTmeasurements using the AUTO method were similar to those derived using the MAN method.  Table 1  Average CIMT measurements using thesemiautomated tracing system (expert reader) Mean CIMT(SD)Max CIMT(SD) Composite 0.707 (0.170) 0.957 (0.226)CCA 0.674 (0.120) 0.878 (0.138)Bulb 0.816 (0.294) 1.149 (0.402)ICA 0.642 (0.253) 0.878 (0.313) CCA  , right common carotid artery;  Bulb  , right carotid bulb;  ICA  , rightinternal carotid artery. Journal of the American Society of Echocardiography  246  Stein et al  March 2005  Intraobserver Reproducibility: Novice Reader  Using the MAN method, the reproducibility of com-posite mean CIMT measurements was worse for theNOV reader (CV     5.2%,  r     .929,  P     .001) thanfor the EXP reader. Using the AUTO method, how-ever, the NOV reader’s reproducibility for compos-ite mean CIMT improved and was quite similar tothat of the EXP reader (CV     3.8%,  r     .959,  P    .001). Similarly, the reproducibility of compositemaximum CIMT measurements by the NOV reader (CV   6.7%,  r   .944,  P   .001) improved using the AUTO method and was very similar to that of theEXP reader (CV   4.0%,  r   .952,  P   .001). Comparison With the Reference Lab: Mean CIMT, Experienced Reader  Using the MAN method, composite mean CIMTreadings by the EXP reader correlated strongly with those of the REF lab (  r     .986,  P     .001), and themean difference of 0.056    0.004 mm was statisti-cally equivalent to that of the REF lab (   P     .001)( Table 3; Figure 2,   A  ). Similar values were obtainedfor the CCA and bulb segments; however, significantdifferences between the EXP reader and the REF labcould not be excluded for the ICA segments. Usingthe AUTO method, composite mean CIMT readingsby the EXP reader correlated strongly (  r   .974,  P   .001). The mean difference of 0.012    0.006 mm was even smaller and was also statistically equivalentto that of the REF lab (   P     .001). Similar meandifferences were obtained for all 6 carotid segments, which were closer to the REF lab readings than tothe MAN readings. Mean CIMT measurements of all6 segments using the AUTO method were statisti-cally equivalent to those from the REF lab. Comparison With the Reference Lab:Maximum CIMT, Experienced Reader  For the EXP reader using the MAN method, compos-ite maximum CIMT readings correlated strongly  with those from the REF lab (  r     .978,  P     .001),and the mean difference of 0.018  0.008 mm wasstatistically equivalent to that of the REF lab (   P    .001) ( Table 3; Figure 2,   B  ). Similar values wereobtained for all 6 carotid segments. Using the AUTOmethod, composite maximum CIMT readings by theEXP reader correlated strongly with those of the REFlab (  r     .967,  P     .001). The mean difference of 0.144    0.009 mm was larger than that observed with the MAN method, but the correlation betweenthe EXP reader and REF lab remained high (  r    0.967), and measurements were statistically equiva-lent to those of the REF lab (   P   .001). Similar values were obtained for all carotid segments except theright bulb, for which significant differences be-tween the EXP reader and the REF lab could not beexcluded. Comparison With the Reference Lab: Mean CIMT, Novice Reader  The NOV reader’s measurements of compositemean CIMT using the MAN method correlatedstrongly (  r   .846,  P   .001) with those of the REFlab and were statistically equivalent (   P     .001)( Table 3; Figure 2,  C   ). However, the NOV reader’sICA and left bulb measurements were thinner thanthose of the REF lab, and significant differencescould not be excluded. Using the AUTO method, thecomposite (   P     .001) and all 6 segmental meanCIMT measurements (   P     .009) of the NOV reader  were statistically equivalent to those of the REF lab. Comparison With the Reference Lab:Maximum CIMT, Novice Reader  Compared with the REF lab, the NOV reader’smeasurements of composite maximum CIMT corre-lated strongly (  r     .827,  P     .001) and werestatistically equivalent (   P  .001) ( Table 3; Figure 2,  D   ). Similar results were noted for all 6 carotidsegments. Using the AUTO method, the composite  Table 2  Intraobserver reproducibility of segmental CIMT measurements (expert reader) Manual Semiautomated Mean     (SEM)  r  (  P ) CV Mean     (SEM)  r  (  P ) CV  Mean CIMTComposite 0.000 (0.004) .986 (  .001) 3.1 0.002 (0.004) .982 (  .001) 3.2CCA 0.000 (0.006) .915 (  .001) 5.4 0.013 (0.004)* .968 (  .001) 3.4Bulb 0.002 (0.008) .980 (  .001) 4.5   0.003 (0.015) .943 (  .001) 6.7ICA 0.004 (0.008) .974 (  .001) 6.8   0.001 (0.008) .981 (  .001) 6.5Maximum CIMTComposite   0.001 (0.010) .950 (  .001) 4.9 0.005 (0.008) .969 (  .001) 4.1CCA 0.034 (0.049) .942 (  .001) 12.4 0.015 (0.007)* .938 (  .001) 3.9Bulb   0.011 (0.015) .960 (  .001) 6.7   0.015 (0.025) .908 (  .001) 9.1ICA    0.013 (0.017) .947 (  .001) 7.7 0.006 (0.012) .967 (  .001) 7.2 CCA  , right common carotid artery;  Bulb  , right carotid bulb;  ICA  , right internal carotid artery.* P   .05 Journal of the American Society of Echocardiography  Volume 18 Number 3  Stein et al  247  maximum CIMT measurements by the NOV reader and the REF lab correlated more closely (  r   .901,  P    .001) and remained statistically equivalent (   P    .006). Using the AUTO method, all segmental max-imum CIMT measurements by the NOV reader werestatistically equivalent except the right bulb (   P    .380). Improvements in Reading Time Using the AUTO Technique The average time for the EXP reader to measure acomplete study was 15.9 (2.3) minutes using theMAN method, but only 10.4 (1.3) minutes using the AUTO method. This average difference of 5.5  0.5minutes represents a 35% reduction in reading time(   P   .001) ( Figure 3 ). For the NOV reader, the average time to measure a complete study decreasedfrom 31.2 (4.0) minutes using the MAN method to16.9 (3.7) minutes using the AUTO method. Thisaverage difference of 14.3  0.9 minutes representsa 46% reduction in reading time (   P   .001). DISCUSSION Carotid artery ultrasound with measurement of CIMT is a safe and noninvasive technique for assess-ing subclinical atherosclerosis and determining car-diovascular risk. 1-5  Although the American Heart Association Prevention Conference V stated that“this test can now be considered. . .at the request of a physician,” CIMT measurements are not widely used in clinical practice, in part because measure-ment of  CIMT requires precision and is time-con-suming. 1  We have developed a novel semiautomatedCIMT border detection program. In the largest vali-dation study of which we are aware, we evaluatedthe reproducibility of its measurements, its accuracy compared with that of a reference laboratory, and itsability to shorten reading time.Our study has several important findings. As dem-onstrated previously, an experienced CIMT reader provided very r epr oducible CIMT measurements by manual tracing. 7,12  Also as expected, the reproduc-ibility of the NOV reader was not as good as that of the EXP reader. Using the AUTO method, however,the NOV reader’s CVs approached those of the EXPreader, indicating that it improved the NOV reader’sreproducibility. This finding is in agreement with aprevious study using a different AUTO method;howev er, it involved fewer carotid arterial seg-ments. 13,14 In our study, mean differences betweenreplicate readings were notably less than or similar to those reported by others using either method,and the CVs for replicate studies using either method also were lower than or similar to thosereported or reviewed in the literature. 7,12-15 The accuracy and reproducibility of the REF lab-oratory has been documented previously. 7,16  Al-though composite and segmental CIMT measure-  Table 3  Comparisons with reference lab for CIMT measurements (read 1  reference lab) Manual Semiautomated Mean     (SEM)  r  (  P correlation  )  P  TOST  Mean     (SEM)  r  (  P correlation  )  P  TOST Expert readerMean CIMTComposite   0.056 (0.004) .986 (  .001)   .001 0.012 (0.006) .974 (  .001)   .001CCA    0.045 (0.002) .923 (  .001)   .001 0.040 (0.007) .931 (  .001)   .001Bulb   0.044 (0.014) .928 (  .001)   .001 0.021 (0.019) .882 (  .001)   .001ICA    0.085 (0.017) .911 (  .001) .092 0.017 (0.012) .956 (  .001)   .001Maximum CIMTComposite 0.018 (0.008) .978 (  .001)   .001 0.144 (0.009) .967 (  .001)   .001CCA 0.012 (0.009) .911 (  .001)   .001 0.164 (0.010) .875 (  .001)   .001Bulb 0.050 (0.023) .885 (  .001)   .001 0.191 (0.030) .863 (  .001) .176ICA    0.005 (0.017) .950 (  .001)   .001 0.118 (0.015) .954 (  .001)   .001Novice readerMean CIMTComposite   0.067 (0.014) .846 (  .001) .003 0.027 (0.009) .943 (  .001)   .001CCA    0.032 (0.007) .925 (  .001)   .001 0.057 (0.007) .920 (  .001)   .001Bulb   0.069 (0.018) .836 (  .001) .016 0.019 (0.016) .877 (  .001)   .001ICA    0.095 (0.019) .884 (  .001) .231   0.007 (0.015) .930 (  .001)   .001Maximum CIMTComposite 0.069 (0.018) .827 (  .001)   .001 0.183 (0.013) .901 (  .001) .006CCA 0.080 (0.008) .891 (  .001)   .001 0.184 (0.009) .906 (  .001)   .001Bulb 0.097 (0.028) .792 (  .001)   .001 0.211 (0.028) .786 (  .001) .380ICA 0.047 (0.020) .909 (  .001)   .001 0.155 (0.021) .919 (  .001) .002 CCA  , right common carotid artery;  Bulb  , right carotid bulb;  ICA  , right internal carotid artery;  TOST  , two one-sided tests. Journal of the American Society of Echocardiography  248  Stein et al  March 2005
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks