| |
Laser scanning as a tool for assessment of HIV-related facial lipoatrophy: evaluation of accuracy and reproducibility
|
| |
| |
HIV Medicine
September 2005
Y Yang and NI Paton
Infectious Disease Research Centre (IDRC), Department of Infectious Diseases, Tan Tock Seng Hospital, Singapore
"...We have developed a standardized method for estimating volume changes in the cheek using changes in facial contour measured by LS, and have shown that the method was accurate in measuring simulated volume change up to 5 mL. The magnitude of facial volume change that occurs in HIV-related LA patients over time has not been specifically quantified in any published study. However, based on magnetic resonance imaging (MRI) measurements of facial fat volume that show a mean difference between patients with severe LA and controls of approximately 15 mL in the cheek [14], and the experience from cosmetic surgery where immediate changes are noticeable in individual subjects after injection of 2-3 mL, volume changes are unlikely to exceed the range that we have tested. In most cases, volume changes are likely to be very small and technically demanding to any technique. LS may be one of the few methods with the precision necessary to detect such small changes.... Although it is unlikely that this technique is going to be useful in the detection and assessment of facial LA over short-term follow-up in individual patients in the clinic setting, with careful standardization of the method it may prove very useful for measuring changes in groups of patients in clinical trials and large cohort studies."
Three-dimensional laser scanning (LS) has been used in research and clinical practice for generating images of facial contour that may be used for planning maxillofacial surgery, measuring craniofacial deformity and planning radiotherapy [7-10]. Laser scanners are easy to use, inexpensive, rapid and noninvasive [7]. This method, with appropriate standardization of scanning technique and image analysis approach, may have potential as an objective tool for the assessment of facial LA.
The aim of this study was to evaluate the accuracy and reproducibility of a standardized method for assessing cheek volume changes using LS, with a view to future application in studies of HIV-related lipodystrophy.
ABSTRACT
Background Facial lipoatrophy (LA) is a common complication of highly active antiretroviral therapy (HAART). Research into causes and treatment of facial LA is hindered by the lack of an objective measurement tool.
Objective To evaluate the accuracy and reproducibility of three-dimensional laser scanning (LS) for estimating cheek volume changes.
Methods Paired laser scans were performed and the images superimposed using commercial software. The volume difference between images was computed within a circle of radius 25 mm placed in a standardized position over the cheek area. Accuracy was tested by scanning before and after known volumes of plasticine (0.5-5 mL) were applied to the cheek area of a mannequin to simulate volume change. Reproducibility was tested by repeated scanning of the mannequin with and without 2 mL of plasticine, and repeated scanning of 10 healthy subjects over the course of 1 week.
Results The mean difference between actual and estimated volume change was small across the range of volumes tested [mean difference 0.08 mL; 95% confidence interval (CI)-0.36 to 0.20 mL). The coefficient of variation for repeated measurements of 2-mL volume change was 5.8%. The intraclass correlation coefficient for scan-to-scan variability was 0.812 (95% CI 0.515-0.947) and for day-to-day variability it was 0.764 (95% CI 0.332-0.935).
Conclusions LS is an accurate and reproducible method for estimating cheek volume changes. It may be useful as an objective tool for assessment of facial LA in clinical research studies.
DISCUSSION
We have developed a standardized method for estimating volume changes in the cheek using changes in facial contour measured by LS, and have shown that the method was accurate in measuring simulated volume change up to 5 mL. The magnitude of facial volume change that occurs in HIV-related LA patients over time has not been specifically quantified in any published study. However, based on magnetic resonance imaging (MRI) measurements of facial fat volume that show a mean difference between patients with severe LA and controls of approximately 15 mL in the cheek [14], and the experience from cosmetic surgery where immediate changes are noticeable in individual subjects after injection of 2-3 mL, volume changes are unlikely to exceed the range that we have tested. In most cases, volume changes are likely to be very small and technically demanding to any technique. LS may be one of the few methods with the precision necessary to detect such small changes.
The method we developed focused on estimating volume change in a defined area of the cheek, as facial appearance suggests that most of the severe changes of LA occur in the cheek area [4,15,16]. The circular area that we have defined might not fully capture all the changes of LA in a particular patient, and thus there may be an underestimate of the actual volume change. However, it is unlikely that changes would occur in other areas without involving the cheek and hence volume change in the cheek is likely to be representative of volume changes elsewhere. Measurement of facial fat volume has shown a similar extent of fat loss in cheek and temporal areas [14]. The use of a standardized area helps to optimize the reproducibility, which would diminish if a larger and less well-defined area were to be used.
In longitudinal studies that attempt to detect small changes, reproducibility of measurements is critical. Variability is a result of a combination of measurement artefact and random biological variation over time. The coefficient of variation of 5.8% for repeated measurements of simulated volume change and the intraclass correlation coefficients of 0.81 and 0.76 for within-day and between-day measurements of human subjects indicate that the method has good reproducibility. These good results are probably attributable in part to the efforts made to standardize the scan conditions, the method of image manipulation for analysis, and the area used for calculation of volume change. The relative contribution of each of these is not known. The only published study in which LS was applied to measurement of lipodystrophy showed slightly greater variability (higher standard deviation) of intermeasurement differences than we found in our study, but the approach used to standardize scan analysis was not reported in sufficient detail to allow speculation on the reasons for the differences [17]. Careful standardization of the position in which the subject is scanned may be particularly important. One study of a portable LS in healthy subjects found good reproducibility (standard deviation for interscan differences of 0.2 mL) when the position of the subject for scanning was carefully standardized, but additional errors of up to 7.6 mL when the subject position was varied [18].
MRI and spiral computer tomography (CT) can provide accurate and direct measurement of facial fat volume in non-HIV-infected patients [19,20]. A study in which volume of facial fat was measured by MRI in HIV-infected patients demonstrated that the method was reproducible and detected the difference in volume between LA and control patients [14]. One study in which single-slice CT was used to measure facial adipose tissue area in HIV-infected patients with LA showed reasonable reproducibility, and that even patients with mild LA could be distinguished from controls [21]. However, these imaging methods are of limited use for repeated measurements in longitudinal clinical studies because of their inconvenience, cost and (in the case of CT) radiation exposure. Sonographic measurement of facial fat at a single point on the cheek has also been shown to be reproducible and able to detect reduced facial fat in patients with LA [22]. The advantages of sonography are that it is rapid, safe and relatively economical, although it is unclear to what extent the results are technique dependent and vary between observers. Similarly, LS also is attractive because of its simplicity and ease of use, and because it offers quick and accurate results with no patient contact or radiation. The combined cost of the scanner and software, currently exceeding US$25 000, may limit widespread use. Although scanning conditions need to be carefully standardized (see above), this is simple to do and does not require specific training. However, some training would be needed for individuals performing scan analysis, but in a research setting the raw data could easily be transmitted for centralized analysis. It is important to note that the method measures change in surface contour, and volume is extrapolated from that. Therefore the results may be affected by other factors that alter facial contour, such as dehydration and variation of human facial expression. The estimated volume change may comprise changes in muscle as well as fat. Furthermore, the performance of the method with the use of facial fillers is uncertain as the contour of these may alter over time without true change in volume, and the method is not suitable for patients with beards or moustaches. Although it is unlikely that this technique is going to be useful in the detection and assessment of facial LA over short-term follow-up in individual patients in the clinic setting, with careful standardization of the method it may prove very useful for measuring changes in groups of patients in clinical trials and large cohort studies.
RESULTS
Accuracy of volume changes
The volume change estimated by LS corresponded closely to the actual volume change across the range of simulated volume changes, with a mean difference of 0.08 mL [95% confidence interval (CI)-0.36 to 0.20 mL]. There was a trend towards an underestimation by LS at higher values of volume change, but the effect was small (regression slope beta=-0.04; P=0.134).
Reproducibility of volume changes
Repeated measurements of the 2-mL volume change gave a mean value of 1.93 mL. The CV was 5.8%.
Reproducibility of measurements in healthy subjects
The reproducibility of the selected areas for volume change estimated within the 25-mm radius circle varied as a result of individual differences in facial contour, but the variation was small (range 1810-1870 mm2). For repeated scans on the same day, the scan-to-scan volume differences from the baseline scan ranged from -0.709 to 0.404 mL. The ICC for the repeated measurements was 0.812 (95% CI 0.515-0.947). For repeated scans over 1 week, the day-to-day volume differences from the baseline scan ranged from -1.760 to 0.817 mL. The ICC was 0.764 (95% CI 0.332-0.935).
METHODS
Fig. 1 Position and size of the plasticine and selected area on the mannequin cheek area.
Volume simulation and human subject reproducibility
The accuracy of the method for estimating volume change was tested using a mannequin (Modern Artificial Model, Modern Artificial Model Pte. Ltd., Singapore). A series of increasing volumes of plasticine (0, 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0, 4.5 and 5.0 mL, TACK-ITtm; A. W. Faber-Castell [M] Sdn. Bhd., Selangor, Malaysia), calculated from known density and weight, was used to simulate changes in facial contour. The plasticine was placed on the cheeks of the mannequin over a circle of 15-mm radius centred over the point one-third of the way along a line drawn between the labial commissure and tragus in order to place it within the area used for volume analysis (see below and Fig. 1). Repeated scans were performed with and without each incremental volume of plasticine.
Reproducibility of the volume change estimates was assessed using five repeated scans with and without a constant 2 mL of plasticine attached to the mannequin cheek. Reproducibility over time was also tested in 10 healthy subjects (five male and five female) recruited from hospital staff with normal facial symmetry and with no beard or moustache. Subjects were scanned a total of eight times over the course of 1 week. Five scans were obtained on the first day to test the scan-to-scan variability, and a single scan was repeated on days 2, 4, and 7 to test day-to-day variability. The study was approved by the local hospital Ethics Committee.
Laser scan procedure
Laser scans were performed using a commercially available three-dimensional digitizer laser scanner (VIVID 300; Minolta Co Ltd, Osaka, Japan). This emits a horizontal stripe light through a cylindrical lens to the subject's face and converts the reflected light into distance information. This process is repeated by scanning the light stripe vertically along the face, and 40 000 points of three-dimensional data are collected during a scan time of 0.6 s.
Before scanning, subjects were asked to remove any jewellery from the face and ears and to moisten and retract any hair to give full exposure of the forehead and ears. They were asked to close their mouth without clenching the teeth, and adopt an expressionless face without smiling. Subjects were seated upright in a standardized swivel chair positioned 120 cm from the laser camera. The swivel chair was initially rotated 45° clockwise from the frontal plane to expose the left cheek, and the subject was instructed to focus their eyes on a marker positioned at an appropriate position on the wall of the room. After obtaining the laser image, the procedure was repeated with the swivel chair rotated to 45° anticlockwise from the frontal plane to obtain an image of the right cheek. An identical procedure was used for the scanning of the mannequin, except that the mannequin was positioned on a box to elevate it within the centre of the scanning gantry.
Analysis of paired laser scans
The three-dimensional scan data were transferred to commercially available computer software for analysis (RapidForm2004; INUS Technology Inc., Seoul, Korea). A pair of scans of the same side of the face was selected, and the 40 000 data points in each scan were automatically converted by the software into a recognizable image of the face for ease of positioning and identification of reference points. The images were cropped to remove the neck region and hair. Four reference points (eye lateral canthus, tragus, labial commissure and ala) that are known to be reproducible were selected on each of the two images [8]. The more recent image was superimposed over the earlier image using the four reference points as a guide. A vector was generated between the midpoint of the bridge of the nose and the chin, and the superimposed image was rotated around this axis to 65° from the frontal view to obtain a standardized image position for analysis. A vector was then drawn between the labial commissure and the tragus, and a point selected one-third of the way along the line from the labial commissure end. A circle of radius 25 mm was positioned with its centre on this point (Fig. 1), and the volume difference between the two components of the superimposed image was calculated within the area of the circle using a customized algorithm developed by the software manufacturer that analyses the distance between the three-dimensional data points underlying the two images. The procedure was repeated for the corresponding images from the opposite cheek and the average of the left and right cheek volume differences was calculated. A single technician performed all the laser scans, identified the soft tissue landmarks, and performed superimpositions and volume change calculations.
Statistical analysis
The accuracy for estimating volume change was assessed by the method of Bland and Altman [11] and linear regression analysis. The reproducibility of the method in the mannequin was determined by calculating the coefficient of variation (CV). CV values less than 10% were considered to represent good reproducibility [12]. The reproducibility of the method in human subjects was assessed by calculating the intraclass correlation coefficient (ICC) for the repeated scans on the first day and for the repeated scans over 1 week. An ICC score of >=0.75 was taken as representing excellent reliability [13]. All P-values were two-sided and P<0.05 was considered statistically significant. The data analyses were performed with SPSS 11.0 software (SPSS, Chicago, IL, USA).
|
|
| |
| |
|
|
|
