RBMOnline - Vol 18. No 5. 2009 658-663 Reproductive BioMedicine Online; www.rbmonline.com/article/3762 on web 26 March 2009 Article Automated follicle tracking improves measurement reliability in patients undergoing ovarian stimulation Nick Raine-Fenning is a consultant gynaecologist and associate professor of reproductive medicine and surgery at the Queen s Medical Centre in Nottingham, UK. He jointly runs the University of Nottingham s assisted conception unit, NURTURE. Nick is an internationally recognised expert in 3-D ultrasound and gynaecological imaging and was awarded a PhD in 2004 for work concerning quantification of pelvic blood flow. He is widely published and is currently lead Editor for gynaecology for the journal Ultrasound in Obstetrics & Gynecology. He is a founder of the recently formed British Society of Gynaecological Imaging, whose aim is to provide access to quality training in gynaecological ultrasound. Dr Nick Raine-Fenning N Raine-Fenning 1,3, K Jayaprakasan 1, S Deb 1, J Clewes 1, I Joergner 2, S Dehghani Bonaki 2, I Johnson 1 1 Nottingham University Research and Treatment Unit in Reproduction, Academic Division of Reproductive Medicine and Surgery, School of Human Development, University of Nottingham, Nottingham, UK; 2 GE Medical Systems, Zipf, Austria 3 Correspondence: e-mail: nick.fenning@nottingham.ac.uk Abstract This study tested the hypothesis that the automated assessment of a stimulated ovary, using 3D ultrasound and sono-avc (automatic volume calculation), provides quicker analysis of follicular number and size than conventional 2D ultrasound, without any loss in measurement validity. Transvaginal ultrasound was performed on day 10 of stimulation in 89 prospectively recruited subjects undergoing IVF treatment. The number and mean diameter of follicles present in both ovaries was measured manually using 2D ultrasound. 3D data were then acquired and analysed using sono-avc. Outcome measures included the number of follicles with a mean diameter >9 mm, >13 mm and >17 mm. The time taken for measurements and data acquisition was recorded. The two methods were compared using a paired t-test or the Wilcoxon signed rank test. Complete data were available for 82 subjects. There was no significant difference in the number of follicles with mean diameters >9 mm, >13 mm and >17 mm measured by either method. The total time taken for follicular measurements was significantly less (P < 0.01) for the automated 3D method (180.5 ± 63.6 versus 236.1 ± 57.1 s) which was associated with significantly less exposure to ultrasound (39.0 ± 6.0 versus 236.10 ± 57.1 s; P < 0.001). Automated 3D follicular measurements using sono- AVC provide a comparable but quicker assessment of follicle number and size. Keywords: follicle tracking, follicle diameter, IVF, sono-avc, three-dimensional ultrasound Introduction 658 A follicle is more likely to contain a mature oocyte if it measures 12 24 mm in diameter (Wittmaack et al., 1994). Oocytes retrieved from follicles measuring 9 15 mm during the follicular phase of the menstrual cycle complete meiotic maturation to metaphase II at a higher rate than oocytes from follicles measuring 3 4 mm in diameter (Tsuji et al., 1985; Whitacre et al., 1998). Follicles as small as 5 7 mm in diameter may be developmentally competent but require maturation in vitro (Wynn et al., 1998; Trounson et al., 2001). Conventional fertility treatment often involves ovulation induction, which is more likely to be successful if the pre-ovulatory follicular diameter is 18 24 mm (Hackeloer et al., 1979) and whilst follicles smaller than 18 mm are capable of producing metaphase II oocytes, fertilization is only 60% of that of the lead follicular group (Rosen et al., 2008). Accurate measurement of follicular size is important and is used to monitor natural cycles and to time the administration of drugs used to promote oocyte maturation prior to intrauterine insemination treatment or oocyte retrieval. Inappropriately early or delayed induction of oocyte maturation may result in an unexpectedly high proportion of immature oocytes or a suboptimal fertilization rate. Whilst there is some degree of flexibility in the timing of oocyte retrieval, particularly 2009 Published by Reproductive Healthcare Ltd, Duck End Farm, Dry Drayton, Cambridge CB23 8DB, UK
when a multi-follicular response has been induced, more valid measurements of follicular size may be expected to be associated with a higher follicle to oocyte ratio and a reduction in the number of immature oocytes, which could result in a higher chance of pregnancy. This may be even more important when minimal approaches to ovarian stimulation are employed or when only a few follicles are developing, as occurs in older women and those with a compromised ovarian reserve. Follicle tracking, which involves the serial assessment of follicle number and size, is commonly employed to assess the response to ovarian stimulation. In the majority of cases the observer uses two-dimensional (2D) ultrasound to identify and then systematically scroll through an ovary, measuring each follicle in turn. There is no consensus as to how many measures should be undertaken or how they are best performed but a single measure is less reliable than two or three measures (Duijkers et al., 2004). During assisted reproduction treatment, using conventional IVF ovarian stimulation protocols, the presence of numerous follicles of different sizes makes such an assessment much harder as the observer has to identify each and every follicle individually and measure it only once. The validity and reliability of such measurements are likely to reduce as the number of follicles increases. Three-dimensional (3D) ultrasound is known to measure volume more accurately than 2D ultrasound and this is also true for the assessment of follicular volume (Kyei-Mensah et al., 1996; Amer et al., 2003). Sono-AVC (automatic volume calculation: GE Medical Systems, Kretz, Austria) is a new software program that identifies and quantifies hypoechoic regions within a 3D dataset and provides an automatic estimation of their absolute dimensions and volume (Raine- Fenning et al., 2007a). Because each different volume is separately colour coded, sono-avc is an ideal tool for studying follicular development in response to ovarian stimulation. Preliminary work has already shown that sono- AVC provides highly reliable and valid measures of follicle diameter and volume (Raine-Fenning et al., 2007a,b, 2008). These studies used sono-avc for the assessment of single follicles to ascertain the reproducibility and accuracy of the software. Assessment of single follicles has limited clinical use in assisted reproduction treatment, however, where ovarian stimulation is generally used to induce a multi-follicular response. The aim of this study was to compare follicle measures made using sono-avc to those made with conventional 2D ultrasound and to assess the time required for measurement. The hypothesis was that the automated measurements would take less time to obtain without affecting their validity. Materials and methods Full ethical approval was given for this study which was conducted in accordance with the ethical principles that have their origin in the Declaration of Helsinki, 1996: the principles of Good Clinical Practice, and the Department of Health Research Governance Framework for Health and Social Care, 2005. Ethical approval was given by the local committee and informed, written consent obtained prior to the enrolment of any subject. A total of 89 consecutive subjects undergoing ovarian stimulation as part of conventional IVF treatment were prospectively recruited. Subjects had to have two ovaries but there were no other exclusion or inclusion criteria. A single observer (JSC) performed transvaginal ultrasound, using a Voluson 730 Expert (GE Medical Systems, Kretz, Austria) and a four-dimensional 5 9 MHz wide band convex volume probe (RIC 5 9), on day 10 of stimulation, when the number and mean diameter of follicles present in both ovaries was measured manually using real-time 2D ultrasound. Each follicle was measured by taking the mean of three planes; the maximal diameter and two planes orthogonal to this, which required manual manipulation of the transducer through 90 degrees. The mean of three measurements was used as this has been shown to be more reliable than a single measure (Duijkers et al., 2004) and is the current clinical practice and standard within the authors assisted conception unit. The time taken for the entire ultrasound examination was recorded to the nearest second. A 3D dataset of each ovary was acquired and these datasets were stored together per subject with a unique identifying study code. The time taken for the ultrasound examination, including the location of each ovary and the subsequent acquisition of 3D data, was recorded as a total for each subject. The 3D ultrasound data were then analysed by the same observer (JSC) at a later time. The order in which the datasets were assessed was determined by a random number generator to ensure the observer was blinded. The two datasets, one for each ovary, were analysed in succession and the combined measures recorded per subject. All measurements were made using the multiplanar view, which provides visualisation of the three orthogonal planes simultaneously and allows the dataset to be manipulated so that the central point of the follicle is consistent for all three images (Figure 1). Once the dataset was correctly positioned, and magnified to provide the best view, the ovary was approximately delineated through the application of a cuboidal region of interest defined using the 3D render mode. This step is necessary as the software only analyses information within this manually defined volume. The software is more likely to incorrectly identify extra-ovarian tissue if an inappropriately large region of interest is selected and will miss follicles if the selected region is too small. Once the volume of interest was selected, sono-avc was activated. Sono-AVC identifies each follicle individually by giving it a specific colour and provides automated measurements of its mean diameter [relaxed sphere diameter, d(v)], its maximum dimensions (dx, dy, dz diameters) and their mean (m-d), and its volume (V) (Figure 2). For the purpose of this study only the relaxed sphere diameter was used for comparison with the real-time 2D measurements. The relaxed sphere diameter was used as this has been shown to provide the most reliable and valid estimate of the true follicular size based on volumetric measurement of the follicular aspirate (Raine-Fenning et al., 2008). A single measurement was obtained in this way, as it has been shown to be sufficiently reliable, and this was used for the comparative analysis. Outcome measures included the time required for assessment and the number of follicles with a mean diameter of 10 mm, 14 mm and 18 mm as these thresholds have clinical importance and affect decision making in the IVF setting (Scott et al., 1989). The time study was conducted according to the technique used and included both the time taken for the ultrasound examination 659
Figure 1. A three-dimensional multiplanar view of a stimulated ovary. The central point of interest can be seen in the longitudinal (upper left image), transverse (upper right image), and coronal plane (lower left image) which can be seen simultaneously providing additional spatial information not available with conventional 2D ultrasound. A three-dimensional rendered image of the whole ovary can be seen in the lower right image. Figure 2. A three-dimensional multiplanar view of a stimulated ovary following the application of Sono-AVC (automatic volume calculation). Each follicle has been individually colour-coded and its dimensions, which have been automatically calculated, are shown on the right of the image as follows; relaxed sphere diameter; d(v), its maximum dimensions (dx, dy, dz) and their mean (m-d), and its volume (V). 660 and follicular measurements. These are presented as a single value for the real-time 2D technique, reflecting clinical practice. This total time was compared with the total time required for the 3D technique. The time required for the 3D technique was also broken down into the time required for actual ultrasound examination (exposure to ultrasound) and data measurement, as the latter was performed as an off-line process and did not require the presence of the subject. Statistical analysis was performed using the Statistical Package for the Social Sciences (SPSS version 14, Chicago, IL, USA). The two methods were compared using a paired t-test or the Wilcoxon signed rank test according to the distribution of the data, assessed using a scatter plot of the difference against the average of the measurements and a histogram of these differences (Bland and Altman, 1986). Limits of agreement were used to examine the validity of measurements according to the absolute difference between the two techniques (Bland and Altman, 1986). Pearson correlation was also used to compare the agreement between the measurement techniques for the number of follicles overall and for the number of follicles within the three different cohorts based on the mean follicle diameter. Results Complete data were available for 82 subjects. Five subjects were excluded as the individual times for the real-time 2D ultrasound assessment or 3D data acquisition were not recorded. Two other subjects were excluded because only one ovarian dataset had been saved. Analysis of these 82 subjects revealed no significant difference in the number of follicles with mean diameters 10 mm, 14
mm and 18 mm measured by the conventional real-time 2D technique or the automated 3D method (Table 1). There was a high degree of correlation between the two measurement methods overall (r = 0.84) and within the different cohorts of follicular size (r = 0.78 for follicles measuring 18 mm, r = 0.87 for follicles measuring 14 mm, and r = 0.96 for follicles measuring 10 mm). The limits of agreement revealed differences between the techniques (Table 2). These were more marked for assessment of the total number of smaller follicles than for the number measuring 18 mm as there were fewer of these. The automated 3D method was associated with a significantly shorter ultrasound examination (P < 0.001) than the conventional 2D technique (Table 3). Additional time was required for data analysis with the 3D technique, as predicted by the study design, but the overall (total) time required for follicle measurements remained significantly less (P < 0.01) than that required for conventional, real-time 2D analysis. The ultrasound examination could be conducted in <1 min in all subjects when 3D was used. The total time required for the assessment of the follicles was approximately 1 min longer with real-time 2D than for the 3D method. Table 1. Comparison of the number of follicles 10 mm, 14 mm and 18 mm measured using sono-avc and real-time 2D ultrasound in 82 subjects undergoing ovarian stimulation as part of conventional assisted reproduction treatment. Measurement Number of Number of Number of technique follicles 10 mm follicles 14 mm follicles 18 mm Sono-AVC 15.90 ± 6.37 (6 26) 8.15 ± 3.68 (2 14) 2.80 ± 1.54 (1 7) Real-time 2D 16.40 ± 6.11 (8 28) 8.65 ± 3.90 (3 17) 2.65 ± 2.01 (1 9) All values expressed as mean ± SD and range. There were no statistically significant differences between the two measurement techniques. Table 2. Inter-method reproducibility of follicle measurements made with sono-avc and real-time 2D ultrasound as assessed by limits of agreement. Measurement Number of Number of Number of technique follicles 10 mm follicles 14 mm follicles 18 mm Mean ± SD 16.15 ± 6.17 8.40 ± 3.66 2.73 ± 1.67 Mean difference ± SD 0.50 ± 1.79 0.50 ± 1.93 0.15 ± 1.27 ULA +3.01 +3.29 +2.64 LLA 4.01 4.29 2.34 ULA, upper limit of agreement; LLA, lower limits of agreement. Table 3. Time required to perform follicle measurements according to measurement technique. 3D ultrasound involves data acquisition and then measurement as two separate processes whereas 2D is a single event. Measurement Time taken for Time taken for Total time taken for technique ultrasound examination (s) measurements (s) assessment (s) 3D Sono-AVC 39.00 ± 6.00 a (26 47) 141.50 ± 64.13 (71 228) 180.50 ± 63.55 b (103 354) Real-time 2D 236.10 ± 57.07 (150 333) 236.10 ± 57.07 a,b (150 333) All values expressed as mean ± SD and range. AVC = automatic volume calculation. a,b Values with the same superscript were significantly different: a, P < 0.001; b, P < 0.01. 661
662 Discussion This is the first study to use sono-avc for the objective assessment of the total follicle population within the whole ovary during ovarian stimulation. The results show the technique provides a comparable estimate of the total number of follicles overall and the number measuring a specific size, compared with conventional real-time 2D ultrasound. The 3D technique significantly reduces the time required for such measurements. The time required with the patient, and their exposure to ultrasound is also significantly reduced, which has important implications for the clinical setting and for bio-safety. Previous studies have shown that the software provides more valid and more reliable assessment of follicular diameter (Raine- Fenning et al., 2007b) and volume (Raine-Fenning et al., 2008) than conventional measures using real-time 2D ultrasound. These preliminary studies were specifically designed to assess the accuracy and reproducibility of the software but did not address the assessment of the whole ovary which is of more clinical relevance. These initial studies were required to ensure the automated measurements were valid and this study was easier to perform with single follicles as it becomes increasingly difficult to reliably identify the same, and therefore the correct, follicle when more are examined simultaneously as has been done in the present study. The authors did not compare the exact measurements of the two techniques for each and every follicle as this is not clinically relevant. Clinical decisions are usually based on the total number of follicles measuring >10 mm and specifically those measuring >17 18 mm as this forms the basis for the timing of oocyte maturation and subsequent egg retrieval (MacDougall et al., 1992; Kolibianakis et al., 2002; Ghosh et al., 2003). The number of follicles measuring >14 mm was also examined in this study as this is the size at which most follicles contain oocytes capable of responding to human chorionic gonadotrophin or recombinant LH given to induce the resumption of meiosis I and germinal vesicle breakdown. The proportion of mature oocytes derived from follicles measuring >15 mm in diameter is relatively constant (Trounson et al., 2001). Significantly fewer mature oocytes are recovered from follicles measuring <15 mm in diameter and only 30% of oocytes will be mature in follicles measuring 12 14 mm. This figure falls to 9% in follicles measuring <11 mm (Scott et al., 1989) and the fertilization rate is considerably lower in oocytes obtained from follicles measuring <10 mm in diameter than in oocytes obtained from larger follicles, which reflects their relative immaturity (Penzias et al., 1994; Wittmaack et al., 1994; Salha et al., 1998). Fertilization rates are higher in follicles measuring >10 mm although it is uncertain whether rates increase progressively after this stage (Haines and Emes, 1991; Penzias et al., 1994; Salha et al., 1998). The degree of accuracy required for follicular assessment is not known and difficult to establish. A mature oocyte can be obtained from a follicle measuring 12 24 mm, although this does not provide any information on its quality or potential. There is a degree of flexibility in the timing of oocyte maturation and a delay of 1 2 days after the leading follicle has reached 18 mm does not appear to result in any significant clinical or embryological differences (MacDougall et al., 1992). The timing of oocyte maturation and collection is based on such measurements, however, and daily ultrasound is often employed towards the end of the follicle stimulation protocol to provide measures of follicle size. Diameter, as a measure, however assumes sphericity and follicles rarely assume such geometry when multifollicular stimulation is induced. Follicles do not actually have a true diameter but this measure is used for convenience. Follicular volume is a more biological measure therefore and is provided by sono-avc. This was not assessed in this study as it was designed to compare the new software with the conventional technique. Limits of agreement were used to evaluate the new automated method to the established real-time 2D technique, as recommended by Bland and Altman (1986). Using this approach, it is possible to detect systematic bias where one of the techniques is constantly obtaining higher values. There were differences between the two techniques in the absolute number of follicles but the significance or importance of these differences is difficult to qualify and open to interpretation as there is no standard or true value. The intraand inter-observer reliability of measures of follicle diameter is better with 3D than real-time 2D ultrasound, regardless of whether manual (Kyei-Mensah et al., 1996) or automated measurements (Raine-Fenning et al., 2008) of the 3D data are made. One could argue, therefore, that the differences relate to a variability or relative inaccuracy in the 2D measures. This assumes the 3D method is the better technique and whilst this is hypothetical it is supported by the previous work that has shown the software provides more valid and more reliable assessment of follicular volume (Raine-Fenning et al., 2008) and diameter (Raine-Fenning et al., 2007b) than 2D ultrasound. Previous work examining the reliability of 2D measures of follicle diameter suggests that repeat measurements may vary by as much as ± 2.4 mm even when three measurements are made and their mean taken, as in this study (Eissa et al., 1985). Volume estimation can be 3.5 ml above or 2.5 ml below the true volume when based on 2D measurements compared with ± 1.0 ml when the volume is predicted from manual 3D measurements (Amer et al., 2003). Automated 3D measurements are more accurate than manually derived ones and can be expected to reduce this error even further (Raine-Fenning et al., 2008). The maximum width for the limits of agreement was not estimated as this study was not designed to calculate how large a difference would be clinically important. This is something that is being examined in a prospective randomized controlled trial where the timing of oocyte maturation and retrieval are being determined by clinicians blinded to the measurement technique used to define follicle number and size. Even if the software does not improve treatment outcome it may still have clinical relevance through its ability to allow offline assessment of 3D data. Patients can be seen more quickly and their examination time reduced, thereby limiting exposure to ultrasound. The stored data can potentially be processed by technicians and incorporated into data archiving and reporting facilities, allowing a virtual real-time reassessment of the ultrasound examination if necessary. In conclusion, automated 3D follicular measurements using sono-avc provide a comparable but quicker measurement of follicular number and size. The duration of the ultrasound examination and exposure of the patient to ultrasound is significantly reduced with the 3D method. This may have important implications in fertility patients and work flow within an assisted conception unit.
Further work is required to see if automated measures made using 3D ultrasound and sono-avc are of biological significance and increase the number of mature oocytes retrieved at egg collection or have any impact on fertilization rates and ultimately pregnancy rates. This study has shown the software has potential and performs at least as well as conventional 2D ultrasound. Acknowledgements Sono-AVC (automatic volume calculation) has been developed by GE Medical Systems in association with K plus Competence Centre (Advanced Computer Vision) and part funded by the K plus Program. References Amer A, Hammadeh ME, Kolkailah M, Ghandour AA 2003 Threedimensional versus two-dimensional ultrasound measurement of follicular volume: are they comparable? Archives of Gynecology and Obstetrics 268, 155 157. Bland JM, Altman DG 1986 Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1, 307 310. Duijkers IJ, Louwe LA, Braat DD, Klipping C 2004 One, two or three: how many directions are useful in transvaginal ultrasound measurement of ovarian follicles? European Journal of Obstetrics, Gynecology and Reproductive Biology 117, 60 63. Eissa MK, Hudson K, Docker MF et al. 1985 Ultrasound follicle diameter measurement: an assessment of interobserver and intraobserver variation. Fertility and Sterility 44, 751 754. Ghosh C, Buck G, Priore R et al. 2003 Follicular response and pregnancy among infertile women undergoing ovulation induction and intrauterine insemination. Fertility and Sterility 80, 328 335. Hackeloer BJ, Fleming R, Robinson HP et al. 1979 Correlation of ultrasonic and endocrinologic assessment of human follicular development. American Journal of Obstetrics and Gynecology 135, 122 128. Haines CJ, Emes AL 1991 The relationship between follicle diameter, fertilization rate, and microscopic embryo quality. Fertility and Sterility 55, 205 207. Kolibianakis E, Bourgain C, Albano C et al. 2002 Effect of ovarian stimulation with recombinant follicle-stimulating hormone, gonadotropin releasing hormone antagonists, and human chorionic gonadotropin on endometrial maturation on the day of oocyte pickup. Fertility and Sterility 78, 1025 1029. Kyei-Mensah A, Zaidi J, Pittrof R et al. 1996 Transvaginal three-dimensional ultrasound: accuracy of follicular volume measurements. Fertility and Sterility 65, 371 376. MacDougall MJ, Tan SL, Jacobs HS 1992 In-vitro fertilization and the ovarian hyperstimulation syndrome. Human Reproduction 7, 597 600. Penzias AS, Emmi AM, Dubey AK et al. 1994 Ultrasound prediction of follicle volume: is the mean diameter reflective? Fertility and Sterility 62, 1274 1276. Raine-Fenning N, Jayaprakasan K, Clewes J et al. 2008 SonoAVC: a novel method of automatic volume calculation. Ultrasound Obstetrics and Gynecology 31, 691 696. Raine-Fenning N, Jayaprakasan K, Clewes J 2007a Automated follicle tracking facilitates standardization and may improve work flow. Ultrasound Obstetrics and Gynecology 30, 1015 1018. Raine-Fenning N, Jayaprakasan K, Clewes J et al. 2007b OC87: the accuracy and variability of manual and automated measurements of follicular diameter. Ultrasound Obstetrics and Gynecology 30, 393 394. Rosen MP, Shen S, Dobson AT et al. 2008 A quantitative assessment of follicle size on oocyte developmental competence. Fertility and Sterility 90, 684 690. Salha O, Nugent D, Dada T et al. 1998 The relationship between follicular fluid aspirate volume and oocyte maturity in in-vitro fertilization cycles. Human Reproduction 13, 1901 1906. Scott RT, Toner JP, Muasher SJ et al. 1989 Follicle-stimulating hormone levels on cycle day 3 are predictive of in vitro fertilization outcome. Fertility and Sterility 51, 651 654. Trounson A, Anderiesz C, Jones G 2001 Maturation of human oocytes in vitro and their developmental competence. Reproduction 121, 51 75. Tsuji K, Sowa M, Nakano R 1985 Relationship between human oocyte maturation and different follicular sizes. Biology of Reproduction 32, 413 417. Whitacre KS, Seifer DB, Friedman CI et al. 1998 Effects of ovarian source, patient age, and menstrual cycle phase on in vitro maturation of immature human oocytes. Fertility and Sterility 70, 1015 1021. Wittmaack FM, Kreger DO, Blasco L et al. 1994 Effect of follicular size on oocyte retrieval, fertilization, cleavage, and embryo quality in in vitro fertilization cycles: a 6-year data collection. Fertility and Sterility 62, 1205 1210. Wynn P, Picton HM, Krapez JA et al. 1998 Pretreatment with follicle stimulating hormone promotes the numbers of human oocytes reaching metaphase II by in-vitro maturation. Human Reproduction 13, 3132 3138. Declaration: The authors report no financial or commercial conflicts of interest. Received 1 July 2008; refereed 26 September 2008; accepted 4 February 2009. 663