Advertisement

Evaluation of oocyte maturity using artificial intelligence quantification of follicle volume biomarker by three-dimensional ultrasound

  • Author Footnotes
    # Contributed equally.
    Xiaowen Liang
    Footnotes
    # Contributed equally.
    Affiliations
    Department of Ultrasound Medicine, the Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China
    Search for articles by this author
  • Author Footnotes
    # Contributed equally.
    Jiamin Liang
    Footnotes
    # Contributed equally.
    Affiliations
    Medical UltraSound Computing (MUSIC) Lab, National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Provincial Key Laboratory of Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Health Science Center, Shenzhen University, Shenzhen, China
    Search for articles by this author
  • Author Footnotes
    # Contributed equally.
    Fengyi Zeng
    Footnotes
    # Contributed equally.
    Affiliations
    Department of Ultrasound Medicine, the Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China
    Search for articles by this author
  • Yan Lin
    Affiliations
    Department of Ultrasound Medicine, the Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China
    Search for articles by this author
  • Yuewei Li
    Affiliations
    Department of Ultrasound Medicine, the Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China
    Search for articles by this author
  • Kuan Cai
    Affiliations
    Department of Ultrasound Medicine, the Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China
    Search for articles by this author
  • Dong Ni
    Affiliations
    Medical UltraSound Computing (MUSIC) Lab, National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Provincial Key Laboratory of Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Health Science Center, Shenzhen University, Shenzhen, China
    Search for articles by this author
  • Zhiyi Chen
    Correspondence
    Corresponding author.
    Affiliations
    Department of Ultrasound Medicine, the Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China

    The Seventh Affiliated Hospital University of South China, Hunan Veterans Administration Hospital, Changsha, China

    The First Affiliated Hospital, Medical Imaging Centre, Hengyang Medical School, University of South China, Hengyang, China

    Institution of Medical Imaging, University of South China, Hengyang, China
    Search for articles by this author
  • Author Footnotes
    # Contributed equally.
Open AccessPublished:July 27, 2022DOI:https://doi.org/10.1016/j.rbmo.2022.07.012

      Abstract

      Research question

      Can a novel deep learning-based follicle volume biomarker using three-dimensional ultrasound (3D-US) be established to aid in the assessment of oocyte maturity, timing of HCG administration and the individual prediction of ovarian hyper-response?

      Design

      A total of 515 IVF cases were enrolled, and 3D-US scanning was carried out on HCG administration day. A follicle volume biomarker established by means of a deep learning-based segmentation algorithm was used to calculate optimal leading follicle volume for predicting number of mature oocytes retrieved and optimizing HCG trigger timing. Performance of the novel biomarker cut-off value was compared with conventional two-dimensional ultrasound (2D-US) follicular diameter measurements in assessing oocyte retrieval outcome. Moreover, demographics, infertility work-up and ultrasound biomarkers were used to build models for predicting ovarian hyper-response.

      Results

      On the basis of the deep learning method, the optimal cut-off value of the follicle volume biomarker was determined to be 0.5 cm3 for predicting number of mature oocytes retrieved; its performance was significantly better than the conventional method (two-dimensional diameter measurement ≥10 mm). The cut-off value for leading follicle volume to optimize HCG trigger timing was determined to be 3.0 cm3 and was significantly associated with a higher number of mature oocytes retrieved (P = 0.01). Accuracy of the multi-layer perceptron model was better than two-dimensional diameter measurement (0.890 versus 0.785) and other multivariate classifiers in predicting ovarian hyper-response (P < 0.001).

      Conclusions

      Deep learning segmentation methods and multivariate classifiers based on 3D-US were found to be potentially effective approaches for assessing mature oocyte retrieval outcome and individual prediction of ovarian hyper-response.

      KEYWORDS

      Introduction

      The aim of assisted reproductive technology (ART) treatment is to achieve and sustain a pregnancy. Exogenous hormones have been used in various forms for hormonal ovarian stimulation in ART with the aim of increasing follicle count (
      • Christianson M.S.
      • Bellver J.
      Innovations in assisted reproductive technologies: impact on contemporary donor egg practice and future advances.
      ). Although mature (metaphase II) oocytes may be obtained from follicles of any size, the mature oocyte retrieval rate was lower in groups with follicles that were too small or too large (
      • Mohr-Sasson A.
      • Orvieto R.
      • Blumenfeld S.
      • Axelrod M.
      • Mor-Hadar D.
      • Grin L.
      • Aizer A.
      • Haas J.
      The association between follicle size and oocyte development as a function of final follicular maturation triggering.
      ). If mature oocyte yield is to be maximized and complications (such as ovarian hyper-stimulation syndrome) are to be avoided, the appropriate assessment of follicular maturation and timing of mature oocyte retrieval is critical (
      • Permadi W.
      • Wahyu Ferdian M.
      • Tjahyadi D.
      • Ardhana Iswari W.
      • Djuwantono T.
      Correlation of Anti-Mullerian Hormone Level and Antral Follicle Count with Oocyte Number in A Fixed-Dose Controlled Ovarian Hyperstimulation of Patients of In Vitro Fertilization Program.
      ). The mean diameter of the leading follicle detected by transvaginal ultrasound has been commonly used to define HCG ovulation trigger injection for many years. Previous studies have indicated that HCG ovulation trigger injection should be timed when the leading follicle reaches 16–22 mm in mean diameter observed by ultrasound (
      • Ovarian Stimulation TEGGO
      • Bosch E.
      • Broer S.
      • Griesinger G.
      • Grynberg M.
      • Humaidan P.
      • Kolibianakis E.
      • Kunicki M.
      • La Marca A.
      • Lainas G.
      • Le Clef N.
      • Massin N.
      • Mastenbroek S.
      • Polyzos N.
      • Sunkara S.K.
      • Timeva T.
      • Töyli M.
      • Urbancsek J.
      • Vermeulen N.
      • Broekmans F.
      ESHRE guideline: ovarian stimulation for IVF/ICSI.
      ;
      • Orvieto R.
      • Mohr-Sasson A.
      • Blumenfeld S.
      • Nahum R.
      • Aizer A.
      • Haas J.
      Does a Large (>24 mm) Follicle Yield a Competent Oocyte/Embryo?.
      ). Another study showed that follicles with a diameter of 12–19 mm at trigger administration were most likely to yield a mature oocyte (
      • Abbara A.
      • Vuong L.N.
      • Ho V.N.A.
      • Clarke S.A.
      • Jeffers L.
      • Comninos A.N.
      • Salim R.
      • Ho T.M.
      • Kelsey T.W.
      • Trew G.H.
      • Humaidan P.
      • Dhillo W.S.
      Follicle Size on Day of Trigger Most Likely to Yield a Mature Oocyte.
      ). In the latest study, showed that punctured follicles measuring 19–24.5 mm in diameter were associated with good-quality blastocysts (
      • Shapiro B.S.
      • Rasouli M.
      • Verma K.
      • Raman A.
      • Garner F.C.
      • Aguirre M.
      • Kaye L.
      • Bedient C.
      The effect of ovarian follicle size on oocyte and embryology outcomes.
      ).
      Tthe number of follicles (mean diameter ≥10 mm) measured manually is usually regarded as the basis for estimating the number of mature oocytes retrieved (
      • Wertheimer A.
      • Nagar R.
      • Oron G.
      • Meizner I.
      • Fisch B.
      • Ben-Haroush A.
      Fertility Treatment Outcomes After Follicle Tracking With Standard 2-Dimensional Sonography Versus 3-Dimensional Sonography-Based Automated Volume Count: Prospective Study.
      ). Nevertheless, several recent studies have questioned the accuracy of the two-dimensional measuring approach mentioned above, especially in multi-follicle development cycles because the shape of follicles is irregular, and mean diameter measured manually by two-dimensional ultrasound (2D-US) neglect the three-dimensional information of follicle shape, which may not reflect the real size of the follicle (
      • Coelho Neto M.A.
      • Ludwin A.
      • Borrell A.
      • Benacerraf B.
      • Dewailly D.
      • da Silva Costa F.
      • Condous G.
      • Alcazar J.L.
      • Jokubkiene L.
      • Guerriero S.
      • Van den Bosch T.
      • Martins W.P.
      Counting ovarian antral follicles by ultrasound: a practical guide.
      ;
      • Re C.
      • Mignini Renzini M.
      • Rodriguez A.
      • Dal Canto M.
      • Buccheri M.
      • Sacchi S.
      • Bartolucci S.
      • Fadini R.
      • La Marca A.
      From a circle to a sphere: the ultrasound imaging of ovarian follicle with 2D and 3D technology.
      ). As such, differences were observed between the number of mature follicles observed by ultrasound and the number of mature oocytes retrieved, which means oocyte maturity may not be optimal at the time of retrieval using the mean diameter for follicles (
      • Liang X.
      • Fang J.
      • Li H.
      • Yang X.
      • Ni D.
      • Zeng F.
      • Chen Z.
      CR-Unet-Based Ultrasonic Follicle Monitoring to Reduce Diameter Variability and Generate Area Automatically as a Novel Biomarker for Follicular Maturity.
      ). Furthermore, manual assessment of follicle number and size is operator-dependent, and inter- and intra-observer variation cannot be ignored (
      • Claman P.
      • Domingo M.
      • Leader A.
      Luteal phase support in in-vitro fertilization using gonadotrophin releasing hormone analogue before ovarian stimulation: a prospective randomized study of human chorionic gonadotrophin versus intramuscular progesterone.
      ;
      • Costello M.F.
      • Shrestha S.M.
      • Sjoblom P.
      • McNally G.
      • Bennett M.J.
      • Steigrad S.J.
      • Hughes G.J.
      Power doppler ultrasound assessment of the relationship between age and ovarian perifollicular blood flow in women undergoing in vitro fertilization treatment.
      ;
      • Xu H.
      • Zeng L.
      • Yang R.
      • Feng Y.
      • Li R.
      • Qiao J.
      Retrospective cohort study: AMH is the best ovarian reserve markers in predicting ovarian response but has unfavorable value in predicting clinical pregnancy in GnRH antagonist protocol.
      ). The advances in three-dimensional ultrasound (3D-US) commercial techniques, i.e. SonoAVC, have made it possible to accurately monitor follicular volumes, but this semi-automatic application is limited by precision of recognition, special instrument requirements and the lack of diagnostic standards (
      • Wertheimer A.
      • Nagar R.
      • Oron G.
      • Meizner I.
      • Fisch B.
      • Ben-Haroush A.
      Fertility Treatment Outcomes After Follicle Tracking With Standard 2-Dimensional Sonography Versus 3-Dimensional Sonography-Based Automated Volume Count: Prospective Study.
      ). To date, methods for optimally selecting suitable patients are lacking, and are, therefore, unable to significantly improve the accuracy of prediction for mature oocytes retrieved by ultrasound (
      • Mascarenhas M.N.
      • Flaxman S.R.
      • Boerma T.
      • Vanderpoel S.
      • Stevens G.A.
      National, regional, and global trends in infertility prevalence since 1990: a systematic analysis of 277 health surveys.
      ).
      Recently, artificial intelligence techniques have emerged as objective, standardized and efficient methods for assessing human reproduction, especially in personalized ovarian stimulation, extended embryo culture, semen analyses, pre-implantation genetic testing and embryo selection (
      • Curchoe C.L.
      • Bormann C.L.
      Artificial intelligence and machine learning for human reproduction and embryology presented at ASRM and ESHRE 2018.
      ;
      • Liu C.
      • Jiao D.
      • Liu Z.
      Artificial Intelligence (AI)-aided Disease Prediction.
      ;
      • Zaninovic N.
      • Rosenwaks Z.
      Artificial intelligence in human in vitro fertilization and embryology.
      ). For the prediction of empty follicles or with oocytes using artificial intelligence techniques, several studies have focused on using the previous ultrasound image segmentation of ovaries and follicles to promote follicular counting accuracy (
      • Li H.
      • Fang J.
      • Liu S.
      • Liang X.
      • Yang X.
      • Mai Z.
      • Van M.T.
      • Wang T.
      • Chen Z.
      • Ni D.
      CR-Unet: A Composite Network for Ovary and Follicle Segmentation in Ultrasound Images.
      ). Few studies, however, have associated the segmentation results with oocyte retrieval outcome. In our previous study, a deep learning framework was proposed for the previous segmentation of 3D-US follicle and ovary images. The dice similarity coefficient (DSC) for follicular segmentation reached 89.11%, which outperformed state-of-the-art methods in various evaluation metrics (
      • Yang X.
      • Li H.
      • Wang Y.
      • Liang X.
      • Chen C.
      • Zhou X.
      • Zeng F.
      • Fang J.
      • Frangi A.
      • Chen Z.
      • Ni D.
      Contrastive rendering with semi-supervised learning for ovary and follicle segmentation from 3D ultrasound.
      ).
      The aim of the present study was to expound an artificial intelligence technique that has the potential to change the management of IVF treatment. The feasibility of two approaches have been discussed to promote the efficiency of oocyte maturity and the timing of oocyte retrieval assessment based on the deep learning segmentation model and machine learning. The relationship between follicle volume biomarker automatically calculated by deep learning models on the day of HCG administration and the number of mature oocytes retrieved was assessed, follicular volume cut-off values that aid in the assessment of oocyte maturity were assessed, and its performance was compared with the two-dimensional diameter measurement. In the meantime, the optimal timing of HCG administration for IVF cycles was determined. The significant features in predicting oocyte maturation, and proposed machine learning methods as novel approaches for predicting ovarian response, were also analysed.

      Materials and methods

      Participants and population

      A total of 515 cases of infertile women undergoing their first ovarian stimulation treatment cycles at the reproductive centre of our hospital between August 2019 and December 2020 were recruited in this prospective study. All patients met the inclusion criteria, as follows: women undergoing ovarian stimulation before IVF using a long protocol, antagonist protocol or mini-stimulation protocol; both ovaries present; and no serious systemic illness. Exclusion criteria were incomplete information or images and abnormal ovarian mass of 3 cm in diameter. Ovarian response was monitored by transvaginal ultrasound and serum oestradiol measures beginning on the fifth day of stimulation. Ovulation was triggered using recombinant HCG (1000 IU or 2000 IU) (Ovitrelle, Merck, Lyon, France) when at least three 17 mm follicles or two 18 mm follicles were observed by ultrasound, and oocyte retrieval was carried out 36 h later. The gonadotrophin releasing hormone antagonist (Cetrorelix) (Merck Serono, Darmstadt, Germany) at 250 mg/day was started on the fifth or sixth day of ovarian stimulation and was continued until the day of HCG administration. Clinical ovarian hyper-response was defined as the retrieval of more than 15 mature oocytes with oestradiol (HCG day) over 5000 ng/l. This study was approved by the local Ethics Committee of the Third Affiliated Hospital of Guangzhou Medical University (number 2018-018), and all patients who agreed to voluntarily participate signed a consent statement.
      A database was created based on the demographics of previous treatment and an infertility work-up, including the cause of infertility, duration of infertility, BMI and hormonal analysis. IVF cycles were also recorded according to stimulation type and dosage, type of trigger and number of mature oocytes retrieved. Patients with any missing data were excluded.

      Ultrasound examination

      Voluson E8 (GE Medical Systems, Zipf, Austria) equipped with a 5- to 9-MHz transvaginal volume probe was used for all ultrasound follicular monitoring. An expert with 8 years’ experience in gynaecologic ultrasound scanned all patients recruited in the study. Each follicle measuring 10 mm or wider in mean diameter was measured by 2D-US, and the number of follicles in both ovaries was recorded, which were based on the sum of follicles in the right and left ovaries. The plane of maximal diameter of the ovary was then revealed, the image stabilized for the three-dimensional volume scanning and was stored as DICOM data. Finally, haemodynamic parameters of the stromal artery were examined, and optimal views were found to obtain a clear and stable doppler flow spectrum. The automatic imputation feature was used to determine peak systolic velocity and resistive index.

      Statistical analyses

      The SPSS version 22.0 package (SPSS, Chicago, IL, USA) was used for statistical analyses. Continuous variables are summarized by mean ± SD. To establish the cut-off value of the follicle volume biomarker that correlates with the outcome of mature oocyte retrieval, the ratio between selected distinct cut-off values and the number of mature oocytes retrieved (ratio = number of mature oocytes retrieved/number of follicles for selected distinct cut-off value) was calculated. That is, the maximum number of follicles at or above the cut-off would be closest to the number of mature oocytes retrieved, and other volume cut-offs either underestimate or overestimate the number of mature oocytes. The Shapiro–Wilk test was used to verify the normality of distribution. Differences in number of mature oocytes among different follicular volume groups were then analysed with either a paired t-test or a Mann–Whitney U test. Homoscedasticity of variance was verified by Levene's test. The comparison between conventional measurement and new follicle volume biomarker was analysed with paired t-tests. P < 0.05 (two-sided test) was considered statistically significant. For a better visualization of results, the average difference of follicle number counting was compared between manual method using ≥ 10 mm two-dimensional diameter measurement and three-dimensional follicle volume biomarker and plotted in a graph (difference = number of mature oocytes retrieved − number of follicle calculated by ultrasound).

      Segmentation model training

      An in-house developed artificial intelligence method based on deep learning was used to obtain the follicle volume biomarker. A detailed technological description of the pipeline has been published separately (
      • Yang X.
      • Li H.
      • Wang Y.
      • Liang X.
      • Chen C.
      • Zhou X.
      • Zeng F.
      • Fang J.
      • Frangi A.
      • Chen Z.
      • Ni D.
      Contrastive rendering with semi-supervised learning for ovary and follicle segmentation from 3D ultrasound.
      ). C-Rend is a deep learning model for the accurate simultaneous segmentation of ovaries and follicles (Figure 1). This novel model overcomes the limitations of SonoAVC, i.e. unavailability of poor-quality images and time consuming image after processing. Three-dimensional U-Net is the backbone structure of C-Rend. It contains the encoder–decoder structure with skip connection, which enables end-to-end training and can directly predict 3D-US segmentation biomarkers of ovaries and follicles. On the basis of the three-dimensional U-Net backbone structure of C-Rend, several changes were made to the training. First, the dataset was enlarged to further improve performance; another independent dataset of 203 total 3D-US volumes from Mindray (Shenzhen, China) and GE (Medical Systems, Zipf, Austria) ultrasound equipment were regarded as our segmentation model dataset, of which 171 volumes were used as a training set and the remaining 32 were used for testing. Second, a larger patch size of 256 × 256 × 256 was used to reduce the ratio of fused follicles. Finally, online data augmentation methods, such as gamma transformation, scaling and random noise adding, were used to improve the robustness of the model.
      Figure 1
      Figure 1The deep learning segmentation framework. Original data of three-dimensional follicle volume biomarkers were calculated by the artificial intelligence method based on deep learning segmentation.

      Prediction model of ovarian hyper-response

      Pearson's correlation coefficient was used to establish the correlation between features and the oocyte retrieval outcome. In the prediction of ovarian hyper-response, univariate analysis was carried out using a threshold classifier to determine the features that had a significant correlation with the outcomes for constructing the multivariate classifier. State-of-the-art machine learning multivariate classifiers were compared in a parallel test, decision tree, support vector machine, k-nearest neighbours, random forest and multi-layer perceptron (MLP). The decision tree used decision rules recursively to partition the feature space. Support vector machine was classified through the decision boundary of the maximum-margin hyperplane. The K-nearest neighbours method predicted discrete class labels based on the nearest patterns. The random forest made the final prediction through multiple decision trees that were randomly chosen. The MLP classifier used the hidden layers containing numbers of nodes, as well as activation function and normalization operation to model the complex relationships among nodes. For each classifier, a grid search for different hyper-parameter values was carried out and the best results were reported. The diagnostic values of single variable and multivariate classifiers were assessed using accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, and area under the curve (AUC). Accuracy indicated the correct classified percentage of all samples. The F1 score considered both positive predictive value and sensitivity.

      Results

      Patient selection

      Complete data were available for 181 patients, and 334 patients were excluded. Patient selection is presented in Figure 2. Baseline characteristics of patients are presented in Table 1.
      Figure 2
      Figure 2Patient selection and study process. 3D-US, three-dimensional ultrasound; IVF-ET, IVF and embryo transfer.
      TABLE 1STUDY PARTICIPANTS
      CharacteristicsValue
      Age, years
       Mean (SD)30.97 (4.02)
       Range2243
      BMI, kg/m2
       Mean (SD)21.76 (3.05)
       Range15.31–31.39
      Cycle day of HCG
       Mean (SD)10.12 (1.98)
       Range5–17
      AMH, ng/ml
       Mean (SD)4.86 (3.87)
       Range0.17–18.00
      LH, IU/l
       Mean (SD)1.70 (1.32)
       Range0.16-9.23
      Progesterone, ng/ml
       Mean (SD)2.73 (1.65)
       Range0.30-15.40
      Oestradiol,
      Oestradiol concentration is formed of endogenous and exogenous hormone. AMH, anti-Müllerian hormone; BMI, body mass index.
      pg/ml
       Mean (SD)12090 (6726.16)
       Range12-46,089
      Number of oocytes retrieved
       Mean (SD)13.97 (7.97)
       Range1-54
      IVF protocol, n (%)
       Long protocol42 (23)
       Antagonist protocol130 (72)
       Mini-stimulation protocol9 (5)
      a Oestradiol concentration is formed of endogenous and exogenous hormone.AMH, anti-Müllerian hormone; BMI, body mass index.

      Statistical analyses

      In this study, data of three-dimensional follicle volume biomarkers were calculated by the artificial intelligence method based on deep learning segmentation. For predicting oocyte maturity, according to the ratio between selected distinct cut-off values and the number of mature oocytes retrieved, the results showed that a follicle volume biomarker 0.5 cm3 or above on the day of HCG administration was more likely to yield mature oocytes, which meant that the number of follicles above this size might correlate best with the number of eggs. The relationship between the number of mature oocytes and different follicular volumes is presented in Table 2. Furthermore, in the comparison of two- and three-dimensional threshold values, when a cut-off value of 0.5 cm3 was chosen, the number of follicles at or above this cut-off was closest to the number of mature oocytes retrieved. Regarding the number of mature oocytes retrieved as the gold standard, the cut-off value of 0.5 cm3 using the three-dimensional follicle volume biomarker performed significantly better than mean diameter measuring 10 mm or wider using the two-dimensional diameter measurement according to the results of a paired t-test (P < 0.001) (Table 2 and Figure 3). Average difference between follicles selected and matures oocytes and its SD using the manual method versus the automatic method were −2.04 ± 0.35 and −0.53 ± 0.31 (P = 0.001), respectively.
      TABLE 2RELATIONSHIP BETWEEN THE NUMBER OF OOCYTES RETRIEVED AND DIFFERENT FOLLICULAR VOLUMES
      Follicle volume/mean diameter value cut-offNumber of mature oocytes retrieved/number of follicles for selected cut-off valueSD
      ≥0.2 cm30.8201.081
      ≥0.3 cm30.8260.236
      ≥0.4 cm30.8980.267
      ≥0.5 cm3,
      Statistically significant.
      0.976
      Statistically significant.
      0.306
      Statistically significant.
      ≥0.6 cm31.0630.351
      ≥0.7 cm31.1360.398
      ≥0.8 cm31.2280.422
      ≥0.9 cm31.3090.459
      ≥1.0 cm31.4180.533
      ≥2.0 cm33.3642.541
      ≥10 mm
      Statistically significant.
      0.870
      Statistically significant.
      0.289
      Statistically significant.
      When a cut-off of 0.5 cm3 is chosen, the maximum number of follicles at or above the cut-off will be closest to the number of mature oocytes retrieved. Other volume cut-offs and mean diameter cut-off (10 mm) either underestimate or overestimate the number of mature oocytes.
      a Statistically significant.
      Figure 3
      Figure 3Difference in follicle number between manual method and automatic method (difference = number of mature oocytes retrieved – number of follicles calculated by ultrasound). In this circumstance, the cut-off of 0.5 cm3 using three-dimensional follicle volume biomarker showed less experimental errors than the cut-off of 10 mm using two-dimensional diameter measurement.
      To determine optimal leading follicle volume to optimize HCG trigger timing, experiments were conducted to explore the optimal threshold and then to validate the findings. All cases were classed by the volume of the leading follicle. In Table 3, Levene's test indicated that all groups met the homogeneity of variance. Shapiro–Wilk tests did not show evidence of normality in all groups. The result showed that a leading follicle over 3.0 cm3 using the follicle volume biomarker as a trigger criterion on HCG administration day was significantly associated with a higher number of mature oocytes retrieved. Differences between class (2, 3) cm3 and class (3, 4) cm3 were significant according to the Mann–Whitney U test (P = 0.02) (Table 3). Therefore, it was concluded that the 3.0 cm3 leading follicle volume cut-off value was significantly associated with a higher number of mature oocytes retrieved. On the basis of the results above, it was further investigated whether a 3.0 cm3 leading follicle volume cut-off value was significantly associated with a higher number of mature oocytes retrieved. Two groups, (0, 3) cm3 and (3, ∞) cm3, were set and significant differences were observed between the two (P = 0.01).
      TABLE 3CORRELATION BETWEEN THE VOLUME OF LEADING FOLLICLE MEASURED BY DEEP LEARNING MODEL AND THE OUTCOME OF MATURE OOCYTES RETRIEVAL
      Groups,
      Data were grouped by volume of the leading follicle.
      cm3
      Cases, nOocytes retrieved, n Mean (SD)P-value (Levene test)P-value (Mann–Whitney U test)
      (0, 2)811.50 (5.98)
      (2, 3)3211.06 (6.99)0.490.44
      (3, 4)7714.26 (7.57)0.940.02
      P < 0.05.
      (4, 5)4114.90 (7.81)0.860.32
      (5, 6)1618.68 (12.50)0.330.19
      (6, 7)1312.69 (5.78)0.280.18
      (7, ∞)518.80 (9.76)0.360.12
      (0, 3)4011.15 (6.73)
      (3, ∞)15214.91 (8.27)0.550.01
      P < 0.05.
      a Data were grouped by volume of the leading follicle.
      b P < 0.05.

      Model evaluation

      In the validation set, the DSC of the improved segmentation deep learning model reached 88.07 for three-dimensional follicular segmentation. In univariate analysis predicting ovarian hyper-response, a total of 26 oocytes retrieval-related characteristics were evaluated (Table 4). These included background information, related hormone parameters, two-dimensional ultrasound parameters (ovarian size, haemodynamic parameters of ovarian stromal artery), total follicular count (calculated manually), number of follicles measuring 10 mm or wider (using the two-dimensional diameter measurement), and number of follicles 0.5 cm3 or above (using the three-dimensional follicle volume biomarker). FSH was excluded as these data were not collected for most patients. Oestradiol mean diameter of ovary, ovarian volume, total follicular count (manually), number of follicles measuring 10 mm or wider (using the two-dimensional diameter measurement) and number of follicles measuring 0.5 cm3 or above (using the three-dimensional follicle volume biomarker) were correlated with the outcome of mature oocyte retrieval. The performance of various characteristics in predicting ovarian hyper-response is presented in Table 4. In contrast to number of follicles (two-dimensional diameter measurement ≥10 mm), number of follicles (three-dimensional follicle volume biomarker ≥0.5 cm3) was a stronger predictor of mature oocytes retrieved (P < 0.001) (Figure 4).
      TABLE 4UNIVARIATE ANALYSIS OF CHARACTERISTICS IN PREDICTING OVARIAN HYPERSTIMULATION
      CharacteristicsACCPPVSENF1SPENPVAUC
      Background information

      Age0.5750.3940.5250.4480.6380.6680.586
      Height0.6190.6690.2040.3020.9190.6200.601
      Weight0.5920.5800.0860.1440.9540.5960.541


      BMI


      0.619


      0.609


      0.246


      0.345


      0.873


      0.620


      0.504
      Infertility duration0.5640.1520.0480.0700.9180.5770.537
      Basic AFC0.6140.5300.5270.5100.6940.6840.706
      Dosage0.6130.5220.5540.5370.6430.6720.606
      Cycle day of HCG0.5860.0000.0000.0001.0000.5860.484
      mparametersAMH0.6470.5650.6530.6000.6540.7320.700
      LH0.5970.4170.1420.1840.9130.6050.570
      Oestradiol0.7730.7460.6830.7090.8380.7870.823
      Progesterone0.6580.6640.3540.4510.8730.6540.670
      Ovarian sizeLong diameter (R)0.6630.6480.5610.5550.7200.7150.768
      Mean diameter (R)0.7240.6870.5500.6030.8300.7360.811
      Short diameter (R)0.7350.7400.5480.6180.8700.7340.812
      Long diameter (L)0.7410.7500.6060.6500.8570.7540.788
      Mean diameter (L)0.7850.7740.6930.7250.8610.7930.800
      Short diameter (L)0.6800.6370.6430.6120.7370.7570.761
      Ovarian volume0.8130.7240.9000.7970.7530.9060.866
      Haemodynamic parameters of ovarian stromal artery

      PSV (R)


      0.559


      0.440


      0.563


      0.468


      0.597


      0.685


      0.680
      PSV (L)0.6200.5310.5480.5150.7020.6990.657
      RI (R)0.6140.7330.0780.1390.9900.6030.496
      RI (L)0.5470.2200.1450.1440.8430.5830.570
      Total follicular count (manually)0.7350.6610.7570.6990.7260.8030.834
      Number of follicles (2DDM ≥10 mm)0.7850.7520.7240.7340.8310.8060.891
      Number of follicles (3DFV ≥0.5 cm3)0.8460.8070.8280.8160.8560.8700.903
      Statistically significant.
      Pearson's correlation was used for statistical analysis. The number of oocytes retrieved was regarded as the gold standard in this study. The number of follicles (3DFV ≥0.5cm3) was significantly associated with the outcome of ovarian hyperstimulation.
      2DDM, two-dimensional diameter measurement; 3DFV, three-dimensional follicle volume biomarker; ACC, accuracy; AUC, area under the curve; F1, F1 score; L, left side; NPV, negative predictive value; PPV, positive predictive value; PSV, peak systolic velocity; R, right side; RI, resistance index; SEN, sensitivity; SPE, specificity.
      a Statistically significant.
      Figure 4
      Figure 4Receiver operator characteristic curves of the two-dimensional diameter measurement and the three-dimensional follicle volume biomarker in predicting the number of mature oocytes retrieved. 2D, two-dimensional; 3D, three-dimensional; FP, false positive; TP, true positive.
      The performance of different multivariate classifiers in predicting ovarian hyper-response is presented in Table 5 and Figure 5. The accuracy of the MLP multivariate classifier was 0.890, which was better than the two-dimensional diameter measurement (0.785) and the single use of the three-dimensional follicle volume biomarker (0.846). According to the result of the t-test, significant differences in the performance of the different models were observed, and MLP with an AUC of 0.880 (95% CI, 0.828 to 0.927) achieved significant improvement over other state-of-the-art methods.
      TABLE 5PERFORMANCE OF MULTIVARIATE CLASSIFIERS IN PREDICTING OVARIAN HYPERSTIMULATION
      ClassifiersACCPPVSENF1SPENPVAUCP-value
      MLP0.8900.9000.8400.8630.9200.8900.880
      DT0.7300.7300.6990.7020.8450.7310.699<0.001
      P < 0.05 according to the result of t-test. ACC, accuracy; AUC, area under the curve; DT, decision tree; F1, F1 score; KNN, k-nearest neighbour; SVM, support vector machine; MLP, multilayer perceptron; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; SEN, sensitivity; SPE, specificity.
      SVM0.8020.8010.7860.7900.8720.7960.786<0.001
      P < 0.05 according to the result of t-test. ACC, accuracy; AUC, area under the curve; DT, decision tree; F1, F1 score; KNN, k-nearest neighbour; SVM, support vector machine; MLP, multilayer perceptron; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; SEN, sensitivity; SPE, specificity.
      KNN0.7470.7410.7300.7300.8370.7520.730<0.001
      P < 0.05 according to the result of t-test. ACC, accuracy; AUC, area under the curve; DT, decision tree; F1, F1 score; KNN, k-nearest neighbour; SVM, support vector machine; MLP, multilayer perceptron; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; SEN, sensitivity; SPE, specificity.
      RF0.8240.8190.8180.8160.8570.8390.8180.008
      P < 0.05 according to the result of t-test. ACC, accuracy; AUC, area under the curve; DT, decision tree; F1, F1 score; KNN, k-nearest neighbour; SVM, support vector machine; MLP, multilayer perceptron; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; SEN, sensitivity; SPE, specificity.
      a P < 0.05 according to the result of t-test.ACC, accuracy; AUC, area under the curve; DT, decision tree; F1, F1 score; KNN, k-nearest neighbour; SVM, support vector machine; MLP, multilayer perceptron; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; SEN, sensitivity; SPE, specificity.
      Figure 5
      Figure 5Receiver operating characteristics curves of multivariate classifiers in predicting ovarian hyper-response. DT, decision tree; KNN, k-nearest neighbour, MLP, multilayer perceptron; RF, random forest; SVM, support vector machine; FP, false positive; TP, true positive.

      Discussion

      This pilot study shows that a prediction of oocyte maturity is achievable by means of a deep learning-based segmentation algorithm and novel follicular volume cut-off values. We further found that multivariate machine learning classifiers represented by MLP achieve great accuracy in the individualized prediction of ovarian hyper-response. The present study has established the premises for additional studies that should determine whether follicle volume biomarker based on artificial intelligence could help improve oocyte retrieval outcome by establishing new criteria for predicting the number of mature oocytes retrieved and determining optimal timing for HCG administration.
      Ultrasonic follicular monitoring is a key step in IVF and ovarian stimulation treatment. In general, the mean diameter and number of follicles measured by 2D-US are used to monitor the development of follicles, predict oocyte maturation and determine the timing of HCG ovulation trigger injection (
      • Noor N.
      • Vignarajan C.P.
      • Malhotra N.
      • Vanamail P.
      Three-Dimensional Automated Volume Calculation (Sonography-Based Automated Volume Count) versus Two-Dimensional Manual Ultrasonography for Follicular Tracking and Oocyte Retrieval in Women Undergoing in vitro Fertilization-Embryo Transfer: A Randomized Controlled Trial.
      ). The 2D-US imaging assumes that follicles have a regular shape, but this is likely inaccurate for follicles during ovarian stimulation because these follicles have uneven and irregular shapes (
      • Revelli A.
      • Martiny G.
      • Delle Piane L.
      • Benedetto C.
      • Rinaudo P.
      Tur-Kaspa I. A critical review of bi-dimensional and three-dimensional ultrasound techniques to monitor follicle growth: do they help improving IVF outcome?.
      ). Recently, 3D-US was proposed as a better solution to improve the evaluation accuracy for ultrasound follicular monitoring, and SonoAVC was the most prominent representative. It was reported that SonoAVC had a good correspondence with two-dimensional follicular measurements and could significantly reduce the time needed to perform ultrasound examination (
      • Rodríguez-Fuentes A.
      • Hernández J.
      • García-Guzman R.
      • Chinea E.
      • Iaconianni L.
      • Palumbo A.
      Prospective evaluation of automated follicle monitoring in 58 in vitro fertilization cycles: follicular volume as a new indicator of oocyte maturity.
      ). In previous studies, however, the limitations of this approach were obvious. A study reported that the process of SonoAVC in 3D-US follicular monitoring was long and semi-automated,and was seriously affected by image quality. Furthermore, results from another study agreed that the use of three-dimensional SonoAVC did not lead to an improvement in cycle parameters (
      • Singh N.
      • Usha B.R.
      • Malik N.
      • Malhotra N.
      • Pant S.
      • Vanamail P.
      Three-dimensional sonography-based automated volume calculation (SonoAVC) versus two-dimensional manual follicular tracking in in vitro fertilization.
      ). Artificial intelligence techniques provide an automated, accurate and convenient path for ultrasound follicular monitoring (
      • Liang X.
      • Fang J.
      • Li H.
      • Yang X.
      • Ni D.
      • Zeng F.
      • Chen Z.
      CR-Unet-Based Ultrasonic Follicle Monitoring to Reduce Diameter Variability and Generate Area Automatically as a Novel Biomarker for Follicular Maturity.
      ). The latest study showed that DSC of 3D-US image segmentation of follicles reached 87% with deep learning algorithms (
      • Mathur P.
      • Kakwani K.
      • Diplav Kudavelly S.
      • Ga R.
      Deep Learning based Quantification of Ovary and Follicles using 3D Transvaginal Ultrasound in Assisted Reproduction.
      ). In this study, a deep learning method was used to improve the accuracy of follicle size measurements, to establish a new criterion for identifying follicles that contain the mature oocyte, and to determine the optimal timing for an ovulation trigger. The results showed that the experimental errors in 88.4% of the cases were acceptable (within six follicles) by the deep learning model. In contrast, only 80.7% of cases were acceptable by two-dimensional diameter measurement (Figure 3). This is a step forward to finding significant differences between two-dimensional and three-dimensional measurements in predicting the outcome of oocyte retrieval.
      In the present study, we proved that a cut-off value for follicle volume biomarker on HCG day is helpful for identifying follicles containing the best oocyte and for determining the optimal time to administer an HCG trigger for final oocyte maturation before ovum retrieval. To the best of our knowledge, this is the first study focu Rodríguez sed on a follicular volume cut-off value to aid in the optimal timing of HCG administration using 3D-US. In predicting maturity of mature oocytes retrieved, it was reported that follicles with a volume of 0.6 cm3 or above on the day of HCG administration (measured by SonoAVC) were more likely to contain mature oocytes (
      • Rodríguez-Fuentes A.
      • Hernández J.
      • García-Guzman R.
      • Chinea E.
      • Iaconianni L.
      • Palumbo A.
      Prospective evaluation of automated follicle monitoring in 58 in vitro fertilization cycles: follicular volume as a new indicator of oocyte maturity.
      ). In another study, follicles in volume range 0.60–0.99 cm3 were also shown to contribute mature oocytes (
      • Hernández J.
      • Rodríguez-Fuentes A.
      • Puopolo M.
      • Palumbo A.
      Follicular Volume Predicts Oocyte Maturity: A Prospective Cohort Study Using Three-Dimensional Ultrasound and SonoAVC.
      ). Some previous studies have compared the true volume of follicular aspirate and follicular volume measured by SonoAVC and concluded that the SonoAVC system slightly underestimated the volume (
      • Raine-Fenning N.
      • Jayaprakasan K.
      • Chamberlain S.
      • Devlin L.
      • Priddle H.
      • Johnson I.
      Automated measurements of follicle diameter: a chance to standardize?.
      ). In our previous study, a deep learning algorithm was shown to have better segmentation performance, especially for small follicles (<5 mm) (
      • Yang X.
      • Li H.
      • Wang Y.
      • Liang X.
      • Chen C.
      • Zhou X.
      • Zeng F.
      • Fang J.
      • Frangi A.
      • Chen Z.
      • Ni D.
      Contrastive rendering with semi-supervised learning for ovary and follicle segmentation from 3D ultrasound.
      ). Interestingly, the results confirmed that follicles with a volume of 0.5 cm3 using the follicle volume biomarker were highly correlated to the number of mature oocytes retrieved. This can be interpreted as the new 3D-US biomarker for oocyte yield prediction using a deep learning segmentation method being more closely representative of actual circumstances.
      Since the outcome of assisted reproductive technology is affected by various related factors, and with factors having individual inaccuracy, the results obtained are often different from the actual results and cannot accurately predict the outcome of oocyte retrieval or pregnancy. In recent years, artificial intelligence technology has been used to combine a variety of fertility-related information and to apply the optimal treatment to improve outcomes, which has helped achieve individual assessment for assisted reproduction (
      • Siristatidis C.
      • Stavros S.
      • Drakeley A.
      • Bettocchi S.
      • Pouliakis A.
      • Drakakis P.
      • Papapanou M.
      • Vlahos N.
      Omics and Artificial Intelligence to Improve In Vitro Fertilization (IVF) Success: A Proposed Protocol.
      ). For example, in the latest retrospective study, a nomogram model was established to predict the probability of ovarian hyperstimulation syndrome in polycystic ovarian syndrome patients, and the AUC was 0.757 (
      • Li F.
      • Chen Y.
      • Niu A.
      • He Y.
      • Yan Y.
      Nomogram Model to Predict the Probability of Ovarian Hyperstimulation Syndrome in the Treatment of Patients With Polycystic Ovary Syndrome.
      ). For the prediction of ovarian hyperstimulation, more than 18 follicles, oestradiol of 5000 ng/l, or both, were regarded as criteria based on the previous guidelines (
      • Humaidan P.
      • Quartarolo J.
      • Papanikolaou E.G.
      Preventing ovarian hyperstimulation syndrome: guidance for the clinician.
      ). Ovarian response is closely correlated to ovarian hyperstimulation. To improve prediction of ovarian hyper-response in each individual patient, we performed univariate analysis and established machine learning classifiers with the new indicator (number of follicles ≥ 0.5 cm3). Our results showed that oestradiol, mean diameter of ovary, ovarian volume, total follicular count (manually), number of follicles measuring 10 mm or wider, and number of follicles measuring 0.5 cm3 or above were correlated with the outcome of oocyte retrieval, which is largely consistent with previous studies (
      • Himabindu Y.
      • Sriharibabu M.
      • Gopinathan K.
      • Satish U.
      • Louis T.F.
      • Gopinath P.
      Anti-mullerian hormone and antral follicle count as predictors of ovarian response in assisted reproduction.
      ;
      • Zheng M.
      • Tong J.
      • Li W.P.
      • Chen Z.J.
      • Zhang C.
      Melatonin concentration in follicular fluid is correlated with antral follicle count (AFC) and in vitro fertilization (IVF) outcomes in women undergoing assisted reproductive technology (ART) procedures.
      ). In addition, we showed that MLP achieved a higher accuracy than both the single use of the follicle volume biomarker and other multivariate classifiers. It can not only accurately predict ovarian hyper-response but can also achieve individual quantitative assessment (AUC = 0.880).
      On the basis of the deep learning segmentation algorithm, the strengths of our study were the establishment of new criteria for predicting oocyte retrieval outcome and the individual prediction of ovarian hyperstimulation. Limitations of our study included the relatively small sample size and the lack of some patient information, i.e. FSH. In addition to the two-dimensional diameter measurement of 10 mm or wider, the results of other two-dimensional thresholds of follicles had not been considered as controls. Furthermore, this was a cohort study, and we did not change the timing of HCG ovulation trigger injection. Novel criteria may be theorized from the results of this study; however, the introduction of new criteria and its clinical application requires a randomized trial demonstrating that improved timing of HCG administration may lead to better fertilization and pregnancy rates. We did not focus on examination time in the study, because the automated approach was not loaded in ultrasound equipment. Also, deep learning typically involves an algorithm task that is specifically trained and improves with continued application. These limitations and further expectations of the present study are aims of ongoing studies.
      In conclusion, artificial intelligence techniques provide a new strategy to solve the problems of available ultrasonic follicular monitoring methods, especially across multiple follicular cycles. Compared with conventional two-dimensional follicular diameter measurements, a cut-off criterion of 3.0 cm3 leading follicle volume for ovulation triggering and 0.5 cm3 leading follicle volume for predicting the number of mature oocytes retrieved using a 3D-US follicle volume biomarker improved oocyte retrieval outcome. Clinical applications of artificial intelligence techniques in the field of reproductive medicine may ultimately encourage doctors to return to the further use of 3D-US in assisted reproduction. Larger prospective randomized controlled trial studies using this methodology are now required.

      Acknowledgements

      All data are available from the corresponding author upon reasonable request. All code is available from the corresponding author upon reasonable request.

      Funding

      This work was supported by Clinical Research 4310 Program of the First Affiliated Hospital of the University of South China (grant number 4310-2021-K06), National Natural Science Foundation of China (grant number 82102054), Major Research Projects of Universities in Guangdong Province (grant number 2019KZDZX1032).

      References

        • Abbara A.
        • Vuong L.N.
        • Ho V.N.A.
        • Clarke S.A.
        • Jeffers L.
        • Comninos A.N.
        • Salim R.
        • Ho T.M.
        • Kelsey T.W.
        • Trew G.H.
        • Humaidan P.
        • Dhillo W.S.
        Follicle Size on Day of Trigger Most Likely to Yield a Mature Oocyte.
        Front Endocrinol. (Lausanne). 2018; 9: 193https://doi.org/10.3389/fendo.2018.00193
        • Christianson M.S.
        • Bellver J.
        Innovations in assisted reproductive technologies: impact on contemporary donor egg practice and future advances.
        Fertil. Steril. 2018; 110: 994-1002https://doi.org/10.1016/j.fertnstert.2018.09.020
        • Claman P.
        • Domingo M.
        • Leader A.
        Luteal phase support in in-vitro fertilization using gonadotrophin releasing hormone analogue before ovarian stimulation: a prospective randomized study of human chorionic gonadotrophin versus intramuscular progesterone.
        Hum. Reprod. 1992; 7: 487-489https://doi.org/10.1093/oxfordjournals.humrep.a137676
        • Coelho Neto M.A.
        • Ludwin A.
        • Borrell A.
        • Benacerraf B.
        • Dewailly D.
        • da Silva Costa F.
        • Condous G.
        • Alcazar J.L.
        • Jokubkiene L.
        • Guerriero S.
        • Van den Bosch T.
        • Martins W.P.
        Counting ovarian antral follicles by ultrasound: a practical guide.
        Ultrasound Obstet. Gynecol. 2018; 51 (Jan): 10-20https://doi.org/10.1002/uog.18945
        • Costello M.F.
        • Shrestha S.M.
        • Sjoblom P.
        • McNally G.
        • Bennett M.J.
        • Steigrad S.J.
        • Hughes G.J.
        Power doppler ultrasound assessment of the relationship between age and ovarian perifollicular blood flow in women undergoing in vitro fertilization treatment.
        J. Assist. Reprod. Genet. 2006; 23: 359-365https://doi.org/10.1007/s10815-006-9067-8
        • Curchoe C.L.
        • Bormann C.L.
        Artificial intelligence and machine learning for human reproduction and embryology presented at ASRM and ESHRE 2018.
        J. Assist. Reprod. Genet. 2019; 36: 591-600https://doi.org/10.1007/s10815-019-01408-x
        • Hernández J.
        • Rodríguez-Fuentes A.
        • Puopolo M.
        • Palumbo A.
        Follicular Volume Predicts Oocyte Maturity: A Prospective Cohort Study Using Three-Dimensional Ultrasound and SonoAVC.
        Reprod. Sci. 2016; 23: 1639-1643https://doi.org/10.1177/1933719116671003
        • Himabindu Y.
        • Sriharibabu M.
        • Gopinathan K.
        • Satish U.
        • Louis T.F.
        • Gopinath P.
        Anti-mullerian hormone and antral follicle count as predictors of ovarian response in assisted reproduction.
        J. Hum. Reprod. Sci. 2013; 6: 27-31https://doi.org/10.4103/0974-1208.112377
        • Humaidan P.
        • Quartarolo J.
        • Papanikolaou E.G.
        Preventing ovarian hyperstimulation syndrome: guidance for the clinician.
        Fertil. Steril. 2010; 94: 389-400https://doi.org/10.1016/j.fertnstert.2010.03.028
        • Li F.
        • Chen Y.
        • Niu A.
        • He Y.
        • Yan Y.
        Nomogram Model to Predict the Probability of Ovarian Hyperstimulation Syndrome in the Treatment of Patients With Polycystic Ovary Syndrome.
        Front Endocrinol. (Lausanne). 2021; 12619059https://doi.org/10.3389/fendo.2021.619059
        • Li H.
        • Fang J.
        • Liu S.
        • Liang X.
        • Yang X.
        • Mai Z.
        • Van M.T.
        • Wang T.
        • Chen Z.
        • Ni D.
        CR-Unet: A Composite Network for Ovary and Follicle Segmentation in Ultrasound Images.
        IEEE J. Biomed. Health Inform. 2020; 24: 974-983https://doi.org/10.1109/JBHI.2019.2946092
        • Liang X.
        • Fang J.
        • Li H.
        • Yang X.
        • Ni D.
        • Zeng F.
        • Chen Z.
        CR-Unet-Based Ultrasonic Follicle Monitoring to Reduce Diameter Variability and Generate Area Automatically as a Novel Biomarker for Follicular Maturity.
        Ultrasound Med. Biol. 2020; 46: 3125-3134https://doi.org/10.1016/j.ultrasmedbio.2020.07.020
        • Liu C.
        • Jiao D.
        • Liu Z.
        Artificial Intelligence (AI)-aided Disease Prediction.
        BIO Integration. 2020; 1: 130-136https://doi.org/10.15212/bioi-2020-0017
        • Mascarenhas M.N.
        • Flaxman S.R.
        • Boerma T.
        • Vanderpoel S.
        • Stevens G.A.
        National, regional, and global trends in infertility prevalence since 1990: a systematic analysis of 277 health surveys.
        PLoS Med. 2012; 9e1001356https://doi.org/10.1371/journal.pmed.1001356
        • Mathur P.
        • Kakwani K.
        • Diplav Kudavelly S.
        • Ga R.
        Deep Learning based Quantification of Ovary and Follicles using 3D Transvaginal Ultrasound in Assisted Reproduction.
        Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. 2020; 2020: 2109-2112https://doi.org/10.1109/EMBC44109.2020.9176703
        • Mohr-Sasson A.
        • Orvieto R.
        • Blumenfeld S.
        • Axelrod M.
        • Mor-Hadar D.
        • Grin L.
        • Aizer A.
        • Haas J.
        The association between follicle size and oocyte development as a function of final follicular maturation triggering.
        Reprod. Biomed. Online. 2020; 40: 887-893https://doi.org/10.1016/j.rbmo.2020.02.005
        • Noor N.
        • Vignarajan C.P.
        • Malhotra N.
        • Vanamail P.
        Three-Dimensional Automated Volume Calculation (Sonography-Based Automated Volume Count) versus Two-Dimensional Manual Ultrasonography for Follicular Tracking and Oocyte Retrieval in Women Undergoing in vitro Fertilization-Embryo Transfer: A Randomized Controlled Trial.
        J. Hum. Reprod. Sci. 2020; 13: 296-302https://doi.org/10.4103/jhrs.JHRS_91_20
        • Orvieto R.
        • Mohr-Sasson A.
        • Blumenfeld S.
        • Nahum R.
        • Aizer A.
        • Haas J.
        Does a Large (>24 mm) Follicle Yield a Competent Oocyte/Embryo?.
        Gynecol. Obstet. Invest. 2020; 85: 416-419https://doi.org/10.1159/000510876
        • Ovarian Stimulation TEGGO
        • Bosch E.
        • Broer S.
        • Griesinger G.
        • Grynberg M.
        • Humaidan P.
        • Kolibianakis E.
        • Kunicki M.
        • La Marca A.
        • Lainas G.
        • Le Clef N.
        • Massin N.
        • Mastenbroek S.
        • Polyzos N.
        • Sunkara S.K.
        • Timeva T.
        • Töyli M.
        • Urbancsek J.
        • Vermeulen N.
        • Broekmans F.
        ESHRE guideline: ovarian stimulation for IVF/ICSI.
        Hum. Reprod. Open. 2020; 2020: hoaa009https://doi.org/10.1093/hropen/hoaa009
        • Permadi W.
        • Wahyu Ferdian M.
        • Tjahyadi D.
        • Ardhana Iswari W.
        • Djuwantono T.
        Correlation of Anti-Mullerian Hormone Level and Antral Follicle Count with Oocyte Number in A Fixed-Dose Controlled Ovarian Hyperstimulation of Patients of In Vitro Fertilization Program.
        Int. J. Fertil. Steril. 2021; 15: 40-43https://doi.org/10.22074/ijfs.2021.6238
        • Raine-Fenning N.
        • Jayaprakasan K.
        • Chamberlain S.
        • Devlin L.
        • Priddle H.
        • Johnson I.
        Automated measurements of follicle diameter: a chance to standardize?.
        Fertil. Steril. 2009; 91: 1469-1472https://doi.org/10.1016/j.fertnstert.2008.07.1719
        • Re C.
        • Mignini Renzini M.
        • Rodriguez A.
        • Dal Canto M.
        • Buccheri M.
        • Sacchi S.
        • Bartolucci S.
        • Fadini R.
        • La Marca A.
        From a circle to a sphere: the ultrasound imaging of ovarian follicle with 2D and 3D technology.
        Gynecol. Endocrinol. 2019; 35: 184-189https://doi.org/10.1080/09513590.2018.1522297
        • Revelli A.
        • Martiny G.
        • Delle Piane L.
        • Benedetto C.
        • Rinaudo P.
        Tur-Kaspa I. A critical review of bi-dimensional and three-dimensional ultrasound techniques to monitor follicle growth: do they help improving IVF outcome?.
        Reprod. Biol. Endocrinol. 2014; 12: 107https://doi.org/10.1186/1477-7827-12-107
        • Rodríguez-Fuentes A.
        • Hernández J.
        • García-Guzman R.
        • Chinea E.
        • Iaconianni L.
        • Palumbo A.
        Prospective evaluation of automated follicle monitoring in 58 in vitro fertilization cycles: follicular volume as a new indicator of oocyte maturity.
        Fertil. Steril. 2010; 93: 616-620https://doi.org/10.1016/j.fertnstert.2009.02.058
        • Shapiro B.S.
        • Rasouli M.
        • Verma K.
        • Raman A.
        • Garner F.C.
        • Aguirre M.
        • Kaye L.
        • Bedient C.
        The effect of ovarian follicle size on oocyte and embryology outcomes.
        Fertil. Steril. 2022; (S0015-0282(22)00136-4)https://doi.org/10.1016/j.fertnstert.2022.02.017
        • Singh N.
        • Usha B.R.
        • Malik N.
        • Malhotra N.
        • Pant S.
        • Vanamail P.
        Three-dimensional sonography-based automated volume calculation (SonoAVC) versus two-dimensional manual follicular tracking in in vitro fertilization.
        Int. J. Gynaecol. Obstet. 2015; 131: 166-169https://doi.org/10.1016/j.ijgo.2015.04.045
        • Siristatidis C.
        • Stavros S.
        • Drakeley A.
        • Bettocchi S.
        • Pouliakis A.
        • Drakakis P.
        • Papapanou M.
        • Vlahos N.
        Omics and Artificial Intelligence to Improve In Vitro Fertilization (IVF) Success: A Proposed Protocol.
        Diagnostics (Basel). 2021; 11: 743https://doi.org/10.3390/diagnostics11050743
        • Wertheimer A.
        • Nagar R.
        • Oron G.
        • Meizner I.
        • Fisch B.
        • Ben-Haroush A.
        Fertility Treatment Outcomes After Follicle Tracking With Standard 2-Dimensional Sonography Versus 3-Dimensional Sonography-Based Automated Volume Count: Prospective Study.
        J. Ultrasound Med. 2018; 37: 859-866https://doi.org/10.1002/jum.14421
        • Xu H.
        • Zeng L.
        • Yang R.
        • Feng Y.
        • Li R.
        • Qiao J.
        Retrospective cohort study: AMH is the best ovarian reserve markers in predicting ovarian response but has unfavorable value in predicting clinical pregnancy in GnRH antagonist protocol.
        Arch. Gynecol. Obstet. 2017; 295: 763-770https://doi.org/10.1007/s00404-016-4274-8
        • Yang X.
        • Li H.
        • Wang Y.
        • Liang X.
        • Chen C.
        • Zhou X.
        • Zeng F.
        • Fang J.
        • Frangi A.
        • Chen Z.
        • Ni D.
        Contrastive rendering with semi-supervised learning for ovary and follicle segmentation from 3D ultrasound.
        Med. Image Anal. 2021; 73102134https://doi.org/10.1016/j.media.2021.102134
        • Zaninovic N.
        • Rosenwaks Z.
        Artificial intelligence in human in vitro fertilization and embryology.
        Fertil. Steril. 2020; 114: 914-920https://doi.org/10.1016/j.fertnstert.2020.09.157
        • Zheng M.
        • Tong J.
        • Li W.P.
        • Chen Z.J.
        • Zhang C.
        Melatonin concentration in follicular fluid is correlated with antral follicle count (AFC) and in vitro fertilization (IVF) outcomes in women undergoing assisted reproductive technology (ART) procedures.
        Gynecol. Endocrinol. 2018; 34: 446-450https://doi.org/10.1080/09513590.2017.1409713

      Biography

      Xiaowen Liang completed her Bachelor of clinical medicine in 2015 and her Master of imaging and nuclear medicine in 2018 at Guangzhou Medical University. She is now working in The Department of Ultrasound Medicine of the Third Affiliated Hospital of Guangzhou Medical University.
      Key message
      Artificial intelligence is being investigated as a promising means for reproductive medicine. A novel biomarker for oocyte maturity evaluation using artificial intelligence-based three-dimensional ultrasound quantification can provide an accurate strategy to solve the problems of available ultrasonic follicular monitoring methods.