P-057 Machine learning-based prediction of testicular sperm extraction: comparison of different preprocessing and models, required sample size and relevance of input biomarkers

Study question

Can advanced machine learning applied to the preoperative assessment predict the testicular sperm extraction outcome in azoospermic context and how many patients are required?

Summary answer

Despite encouraging results (AUC = 92.0%, sensitivity = 83.9% and specificity = 84.2%), integrating new biomarkers would probably be more relevant than enrolling additional patients.

What is known already

Testicular sperm extraction (TESE) is an essential therapeutic tool for the male infertility management and is often the “last hope” before gamete donation for these patients. However, it is an invasive procedure and is successful in up to 50%. Until now, no model is sufficiently powerful to accurately predict the success of sperm retrieval in TESE. Among the few models already developed, the findings are highly disparate despite having common input data (preoperative assessment). Moreover, only few types of machine learning models and procedures have been investigated. Performances were mostly capped despite the inclusion sometimes of more than 1000 patients.

Study design, size, duration

Data of 175 patients who underwent TESE between 2012 and 2021 were retrospectively analyzed. The performances of a wide range of preprocessing methods and machine learning models (state-of-the-art methods in machine learning) we explored, evaluated, and compared. The objective was to predict the presence or absence of spermatozoa, using 17 parameters (clinical, hormonal, genetic, history) from the preoperative assessment. The study protocol was approved by a local ethics committee (IRB CER-2021-041).

Participants/materials, setting, methods

After data preprocessing (standardization…), Machine Learning models (Bayesian Naive Classification, logistic regression, k-nearest neighbor classifier, support vector machine, random forests, GradientBoosting and XGBoost) and Deep Learning models were tested. The validation procedure consisted of splitting the dataset into a training set and test set. Beyond the standard metrics (sensitivity, specificity, AUC-ROC), the identification of the most relevant variables and the learning curve to determine the optimal patient number to be included were performed.

Main results and the role of chance

At least one live spermatozoon was found in the testicular tissue of 104 (59.4%) patients (positive TESE) out of 175. The best performing model (Random Forest with appropriate preprocessing) obtained the following results on the test set: AUC = 92.0%, sensitivity = 83.9% and specificity = 84.2%, leading to an efficient tool, which gives additional and more relevant information than the different variables taken separately. Inhibin B, FSH and history of cryptorchidism were the variables with the most discriminating power. However, a plateau in the model performance was observed (beyond 110 patients), whatever the approach or the preprocessing used. A trend curve shows that beyond 110 patients, no improvement can be observed and cast doubt about the power of the traditional preoperative parameters assessed before TESE. The classic preoperative assessment can probably not fully predict the TESE outcomes. Further work is needed to be enhance with new hypothesis and the use of new biomarkers to be integrated into the models.

Limitations, reasons for caution

The main limitation was the monocentric design and the use of retrospective data.

Wider implications of the findings

Machine learning models can provide the basis for an enhanced decision support system tool in the context of azoospermia. Indefinitely increasing the number of participants is not likely to be the solution: further hypotheses and biomarkers integration into the models will probably be necessary to improve performance.

Trial registration number

not applicable

The impact of optimal dating on the assessment of fetal growth

BACKGROUND

The impact of using the Intergrowth (IG) dating formulae in comparison to the commonly used Robinson dating on the evaluation of biometrics and estimated fetal weight (EFW) has not been evaluated.

METHODS

Nationwide cross-sectional study of routine fetal ultrasound biometry in low-risk pregnant women whose gestational age (GA) had been previously assessed by a first trimester CRL measurement. We compared the CRL-based GA according to the Robinson formula and the IG formula. We evaluated the fetal biometric measurements as well as the EFW taken later in pregnancy depending on the dating formula used. Mean and standard deviation of the Z scores as well as the number and percentage of cases classified as <3rd, < 10th, >90th and > 97th percentile were compared.

RESULTS

Three thousand five hundred twenty-two low-risk women with scans carried out after 18 weeks were included. There were differences of zero, one and 2 days in 642 (18.2%), 2700 (76.7%) and 180 (5%) when GA was estimated based on the Robinson or the IG formula, respectively. The biometry Z scores assessed later in pregnancy were all statistically significantly lower when the Intergrowth-based dating formula was used (p < 10- 4). Likewise, the number and percentage of foetuses classified as <3rd, < 10th, >90th and > 97th percentile demonstrated significant differences. As an example, the proportion of SGA foetuses varied from 3.46 to 4.57% (p = 0.02) and that of LGA foetuses from 17.86 to 13.4% (p < 10- 4).

CONCLUSION

The dating formula used has a quite significant impact on the subsequent evaluation of biometry and EFW. We suggest that the combined and homogeneous use of a recent dating standard, together with prescriptive growth standards established on the same low-risk pregnancies, allows an optimal assessment of fetal growth.

New insights in cerebral findings associated with fetal myelomeningocele: a retrospective cohort study in a single tertiary centre

OBJECTIVE

To investigate cerebral anomalies other than Chiari type 2 malformation in fetuses with myelomeningocele (MMC).

DESIGN

A retrospective cohort study in a single tertiary centre.

SETTING

A review of associated cerebral anomalies in cases with prenatal diagnosis of myelomeningocele.

POPULATION

Seventy cases of fetal myelomeningocele.

METHODS

Ultrasound and MRI images were blindly reviewed. Postnatal imaging and results of the postmortem results were also reviewed. The association between cerebral anomalies and the following ultrasound findings was measured: level of the defect, ventriculomegaly, microcephaly and fetal talipes.

MAIN OUTCOME MEASURES

A microcephaly was observed in 32/70 cases (46%) and a ventriculomegaly was observed in 39/70 cases (56%). Other cerebral anomalies were diagnosed in 47/70 (67%).

RESULTS

Other cerebral anomalies were represented by 42/70 cases with abnormal CC (60%), 8/70 cases with perinodular heterotopia (PNH; 11%), 2/70 cases with abnormal gyration (3%). MRI performed only in fetal surgery cases confirmed the ulltrasound findings in all cases and provided additional findings in two cases (PNH). Risk ratios of fetal cerebral anomalies associated with MMC did not reach significance for microcephaly, ventriculomegaly, talipes or the level of the defect There was an overall good correlation between pre- and postnatal findings with a Kappa value of 0.79 [95% CI 0.57-1] and 82% agreement.

CONCLUSION

Fetal brain anomalies other than Chiari type 2 malformation are frequently observed in fetuses with myelomeningocele, predominantly represented by CC anomalies. Whether these associated cerebral anomalies have an impact on selecting cases eligible for fetal surgery needs further evaluation.

TWEETABLE ABSTRACT

Fetal cerebral anomalies other than Chiari type 2 malformation, microcephaly, and ventriculomegaly may be associated with MMC in up to 67% of the cases.

[Feticide procedures in second and third trimesters terminations of pregnancy]

Performing a feticide as part of termination of late pregnancy is recommended in many countries. Feticide avoids a live birth of a severely affected premature newborn and prevents fetal pain. There are limited data on feticide procedures since only a few countries in the world authorize late termination of pregnancy. The objective of this review was to assess the most appropriate feticide procedure based on published data during the last thirty years. Administration of an initial fetal analgesia followed by a lethal lidocaine injection through the umbilical cord, under ultrasound guidance, appears to be the most effective, safe and ethical way to perform feticide. According to the current knowledge regarding the risk of fetal pain and survival of extremely preterm infants, a feticide should be discussed as early as 20-22 weeks of gestation.

[Prenatal diagnosis and postnatal outcome of isolated intra-abdominal calcifications: A 10-year experience from a referral fetal medicine center]

INTRODUCTION

Intra-abdominal calcifications (iAC) detected during fetal ultrasound examinations are characterized by their isolated or associated nature, as well as their location. Our objective was to describe all cases of isolated iAC along with their etiological investigations and neonatal outcome, during a 10-year practice in a referral center.

METHODS

We conducted a retrospective descriptive monocentric study on neonates diagnosed with isolated iAC after antenatal expert ultrasound scan and referred to the Multidisciplinary Center for Prenatal Diagnosis at Trousseau Hospital and born between January 1st, 2008 and June 30th, 2018. The exclusion criteria were: retroperitoneal calcifications, iAC associated with other digestive abnormalities or with congenital malformations.

RESULTS

The 32 isolated iAC cases accounted for 46% of all iAC. Nine cases were excluded for missing neonatal data. Among the 23 remaining isolated iAC cases, we observed 15 intra-hepatic calcifications, 5 peri-hepatic and two peritoneal calcifications. One fetus had both intra- and peri-hepatic calcifications. The majority of iAC remained stable throughout pregnancy. No cases of aneuploidy, fetal infection, or cystic fibrosis were detected. The neonatal outcome was favorable in all cases.

CONCLUSIONS

In case of isolated and stable iAC after expert ultrasound scan, after having ruled out infectious diseases of the fetus and looked for the most frequent mutations of cystic fibrosis in the parents, the prognosis is favorable. Fetal karyotyping is recommended when additional structural anomalies are present.

Evaluation of septal insertion of atrioventricular valves in fetuses by postmortem 4.7 Tesla cardiac MRI: A feasibility study

PURPOSE

The purpose of this study was to compare non-invasive high-spatial-resolution postmortem cardiac magnetic resonance imaging (MRI) and autopsy findings for evaluating the septal insertion of atrioventricular valves in fetuses.

MATERIALS AND METHODS

Five fetal heart specimens including two normal hearts, one heart with complete atrioventricular septal defect (AVSD) and two hearts with linear insertion of atrioventricular valves (LIAVV; gestational age 17 to 34 weeks) were studied with cardiac MRI using a 4.7 T MRI scanner without sample preparation. Three (3D) and two-dimensional (2D) turbo-RARE (rapid imaging with refocused echoes) sequences in four-chamber and left-ventricular long-axis planes were obtained with a minimal isotropic/in-plane resolution of 156μm. Nonparametric tests were performed to compare the distance between insertions of medial leaflets of the atrioventricular valves and the inlet/outlet distance ratio between MRI and autopsy findings in normal, complete AVSD and with linear insertion of atrioventricular valves (LIAVV) fetal hearts.

RESULTS

Despite apparent differences between LIAVV/normal hearts, no significant differences were found between differential insertion of medial leaflets and inlet/outlet distance ratios with both techniques. Very good to excellent reliability between both techniques was found for differential insertion (ICC: 87.2%; 95% CI: -21.7%, 99.1%) (P=0.963) and inlet/outlet distance ratio (ICC 98.3%; 95%CI: 85.2%, 99.8%) (P=0.537) measurements.

CONCLUSION

Postmortem cardiac MRI could replace autopsy for assessing normal or abnormal septal insertion of atrioventricular valves in fetuses without requiring specific preparation of the heart.

Knowledge Representation and Management, It's Time to Integrate!

Objectives: To select, present, and summarize the best papers published in 2016 in the field of Knowledge Representation and Management (KRM). Methods: A comprehensive and standardized review of the medical informatics literature was performed based on a PubMed query. Results: Among the 1,421 retrieved papers, the review process resulted in the selection of four best papers focused on the integration of heterogeneous data via the development and the alignment of terminological resources. In the first article, the authors provide a curated and standardized version of the publicly available US FDA Adverse Event Reporting System. Such a resource will improve the quality of the underlying data, and enable standardized analyses using common vocabularies. The second article describes a project developed in order to facilitate heterogeneous data integration in the i2b2 framework. The originality is to allow users integrate the data described in different terminologies and to build a new repository, with a unique model able to support the representation of the various data. The third paper is dedicated to model the association between multiple phenotypic traits described within the Human Phenotype Ontology (HPO) and the corresponding genotype in the specific context of rare diseases (rare variants). Finally, the fourth paper presents solutions to annotation-ontology mapping in genome-scale data. Of particular interest in this work is the Experimental Factor Ontology (EFO) and its generic association model, the Ontology of Biomedical AssociatioN (OBAN). Conclusion: Ontologies have started to show their efficiency to integrate medical data for various tasks in medical informatics: electronic health records data management, clinical research, and knowledge-based systems development.

[Ultrasound screening for birth defects: A medico-economic review]

OBJECTIVES

The systematic use of ultrasound during pregnancy aims at birth defect detection. Our objective was to assess the economic efficiency of prenatal ultrasound screening for fetal malformations.

METHODS

We carried out a literature review on Medline via PubMed between 1985 and 2015, from the economic perspective of the prenatal ultrasound screening for fetal malformations.

RESULTS

The literature on this subject was sparse and we selected only twelve articles presenting relevant economic data, of which only eight were proper medico-economic studies. We found arguments for the economic effectiveness of ultrasound screening for fetal malformation detection, which is largely linked to the terminations of pregnancies and to the cost of the handicaps "avoided". However, none of the reviewed articles could reach medico-economic conclusions. Additionally, we highlighted various elements making economic analyses more complex in this field: the choice of the method, the uncertainty around two essential parameters (the efficiency of ultrasound and the costs of procedures) and the difficulties to compare or to generalize results. We also noticed important methodological heterogeneity among the studies and the absence of French study.

CONCLUSIONS

Previously published data are insufficient to assess the economic efficiency of prenatal ultrasound screening for fetal malformations.

Towards ontology-based decision support systems for complex ultrasound diagnosis in obstetrics and gynecology

INTRODUCTION

We have developed a new knowledge base intelligent system for obstetrics and gynecology ultrasound imaging, based on an ontology and a reference image collection. This study evaluates the new system to support accurate annotations of ultrasound images. We have used the early ultrasound diagnosis of ectopic pregnancies as a model clinical issue.

MATERIAL AND METHODS

The ectopic pregnancy ontology was derived from medical texts (4260 ultrasound reports of ectopic pregnancy from a specialist center in the UK and 2795 Pubmed abstracts indexed with the MeSH term "Pregnancy, Ectopic") and the reference image collection was built on a selection from 106 publications. We conducted a retrospective analysis of the signs in 35 scans of ectopic pregnancy by six observers using the new system.

RESULTS

The resulting ectopic pregnancy ontology consisted of 1395 terms, and 80 images were collected for the reference collection. The observers used the knowledge base intelligent system to provide a total of 1486 sign annotations. The precision, recall and F-measure for the annotations were 0.83, 0.62 and 0.71, respectively. The global proportion of agreement was 40.35% 95% CI [38.64-42.05].

DISCUSSION

The ontology-based intelligent system provides accurate annotations of ultrasound images and suggests that it may benefit non-expert operators. The precision rate is appropriate for accurate input of a computer-based clinical decision support and could be used to support medical imaging diagnosis of complex conditions in obstetrics and gynecology.

[Fetal abdominal cysts at the first trimester scan]

OBJECTIVES

Abdominal cysts are seldom detected at the first trimester scan. The aim of this study is to ascertain their outcome, which is currently not established.

METHODS

The French College of Fetal Ultrasound conducted a prospective observational study of 24months, collecting all cases of abdominal cysts discovered during the first trimester ultrasound. Cases of megacystis were excluded from the study. Ultrasound images, prenatal diagnosis expert reports and pregnancy outcomes were collected by sonographers after patient consent.

RESULTS

Ten cases of abdominal cysts were collected. The cysts had a mean diameter of 15mm. They were anechoic in 5 cases, hyperechoic in 2 cases and mixed in 3 cases. In 6 of 10 cases, complete resolution was observed at 18WG with a good post-natal outcome; the five cases with anechoic images were associated with normal pediatric examination at birth and in the case of the resolved hyperechoic image, an isolated imperforate anus was observed. In the four cases of hyperechoic or mixed images that had not resolved, the outcome was poor with four termination of pregnancies, including two cases of cloacal dysgenesis.

CONCLUSIONS

Abdominal cysts are rare at the first trimester scan. They resolve in two thirds of cases and are then associated with good outcome. When they do not resolve or when they are not strictly anechoic, they require a referral ultrasound examination at 18 and 22WG.

Searching for rare diseases in PubMed: a blind comparison of Orphanet expert query and query based on terminological knowledge

BACKGROUND

Despite international initiatives like Orphanet, it remains difficult to find up-to-date information about rare diseases. The aim of this study is to propose an exhaustive set of queries for PubMed based on terminological knowledge and to evaluate it versus the queries based on expertise provided by the most frequently used resource in Europe: Orphanet.

METHODS

Four rare disease terminologies (MeSH, OMIM, HPO and HRDO) were manually mapped to each other permitting the automatic creation of expended terminological queries for rare diseases. For 30 rare diseases, 30 citations retrieved by Orphanet expert query and/or query based on terminological knowledge were assessed for relevance by two independent reviewers unaware of the query's origin. An adjudication procedure was used to resolve any discrepancy. Precision, relative recall and F-measure were all computed.

RESULTS

For each Orphanet rare disease (n = 8982), there was a corresponding terminological query, in contrast with only 2284 queries provided by Orphanet. Only 553 citations were evaluated due to queries with 0 or only a few hits. There were no significant differences between the Orpha query and terminological query in terms of precision, respectively 0.61 vs 0.52 (p = 0.13). Nevertheless, terminological queries retrieved more citations more often than Orpha queries (0.57 vs. 0.33; p = 0.01). Interestingly, Orpha queries seemed to retrieve older citations than terminological queries (p < 0.0001).

CONCLUSION

The terminological queries proposed in this study are now currently available for all rare diseases. They may be a useful tool for both precision or recall oriented literature search.

[How to assess the neutral position of the fetus for the crown-rump length measurement at the nuchal translucency scan]

OBJECTIVES

The objective of this study was to establish a simple and reproducible method for the assessment of the fetal head position when measuring crown-rump length (CRL) at the nuchal translucency scan.

METHODS

Two observers conducted a retrospective analysis of a consecutive series of 570 images of CRL collected by the French College of Fetal Echography (CFEF) national practice assessment program for the first-trimester scan. The images were deemed hyper-flexed if no fluid was visible between the chin and the chest of the fetus. The images were deemed hyper-extended if the angle between the palate and the CRL line was 90° or more. The images were deemed neutral if no hyper-extension nor hyper-flexion was observed.

RESULTS

The proportion of agreement for a non-neutral position of the fetal head was 91.3% (kappa=0.80, 95% CI [0.75 to 0.86]). Images with a non-neutral position corresponded to poor CRL quality images according to the CFEF score (relative risk=4.2, 95% CI [2.9 to 6.1] for one observer and 4.9, 95% CI [3.3 to 7.2] for the other observer). Proportions of agreement for the hyper-flexion and for the hyper-extension were 94.6% (kappa=0.80, 95% CI [0.72 to 0.87]) and 96.8% (kappa=0.81, 95% CI [0.72 to 0.90]), respectively. No additional lines were drawn on the ultrasound images during the evaluation process.

CONCLUSIONS

The proposed method for evaluating the extent to which the fetal head has a non-neutral position at measurement of the CRL appears both simple and reproducible.

[A simple self-assessment tool for the first-trimester ultrasound images]

OBJECTIVES

The objective of this study was to establish a minimum subset of simple criteria for the self-assessment of the quality of first-trimester ultrasound images of nuchal translucency (NT) and crown-rump length (CRL).

METHODS

We designed 162 simplified image-scoring methods (ISM) based on 1, 2, 3 or 4 binary criteria derived from the 8 criteria of the original image scoring method of the French College of Fetal Ultrasound (CFEF). These ISM were assessed on 68,250 consecutive scans of the French national audit conducted by the CFEF on NT and CRL images. The ISM associated with the best precision to identify excellent/reasonable quality scans were selected.

RESULTS

Simplified ISM based on 1, 2, 3 and 4 criteria showed maximum positive predictive values of 95.3% (95.11-95.50) 98.0% (97.87-98.14), 99.3% (99.17-99.35) and 99.7% (99.68-99.79), respectively, to identify excellent/reasonable quality scans. The proportion of excellent/reasonable scans was 2.8 to 16.7% when three criteria among the 8 were insufficient, and 0.17 to 3.95% when four criteria were insufficient.

CONCLUSIONS

The best performing ISM was based on the following four quality criteria: (i) sagittal plane of the NT, (ii) calipers placement for measuring the NT, (iii) image magnification of NT images and (iv) CRL measurement. This score might be the most relevant in clinical practice in the first-trimester screening.

Screening for fetal spina bifida at the 11-13-week scan using three anatomical features of the posterior brain

OBJECTIVE

To evaluate the contribution of examination of specific anatomical features of the fetal posterior brain on mid-sagittal first-trimester ultrasound examination to the early detection of open spina bifida.

METHODS

Four independent observers reviewed a series of 260 mid-sagittal first-trimester ultrasound images from 52 cases of open spina bifida and 208 normal fetuses. The following analysis was performed by each reviewer for each image: Herman score calculation, intracranial translucency score (CFEF-IT) calculation and determination of presence or absence of three anatomical criteria: intracranial translucency (IT), caudal displacement of the brainstem and cisterna magna. The sensitivity and the false-positive rate for spina bifida detection were calculated for each of the latter three criteria. A secondary analysis was performed on the subset of images achieving a Herman score ≥ 7.

RESULTS

The highest detection rate for spina bifida was achieved by non-visualization of the cisterna magna, with associated sensitivity of 50-73% and 39-76%, respectively, for all images and for the subset of images achieving a Herman score ≥ 7. Posterior shift of the brainstem achieved the highest detection rate (86%), but for a single reviewer only. The level of variation in performance between observers was also greatest for this sign. Absence of IT was associated with a lower detection rate for all observers. Overall, an abnormal posterior brain presenting at least one of these three criteria was associated with a detection rate ranging from 50 to 90%.

CONCLUSION

In the detection of spina bifida, non-visualization of the cisterna magna achieved the best screening performance. Both non-visualization of the IT and posterior shift of the brainstem were associated with acceptable but lower detection rates. A prospective evaluation of changes in the posterior brain is needed to allow assessment of the most pertinent criteria for first-trimester screening for spina bifida.

Rare diseases knowledge management: the contribution of proximity measurements in OntoOrpha and OMIM

In this paper, we introduce an application of Proxima and define a new measure of proximity between two concepts present in an ontology. The approach is based on the three dimensions of a conceptualization: intention with relations between concepts, expression with terms denoting concepts, and extension with instances of concepts. This preliminary work, in the field of rare diseases, involved the Orphanet Ontology of Rare Diseases (OntoOrpha) and corpus of texts extracted from Online Inheritance in Man (OMIM). The proximity measurements are consistent with an appropriate representation of groups of diseases in the ontology, which are derived from the Orphanet classifications of rare diseases. Other semantic relations are explored and new perspectives in medical knowledge curation are proposed.

Prenatal ultrasonographic diagnosis of polymicrogyria

We report a rare case of polymicrogyria diagnosed at 27 weeks' gestation on ultrasound examination and associated with cytomegalovirus (CMV) infection. The ultrasound finding suggesting this diagnosis was the direct visibility of the overfolded cortical ribbon. The cerebral surface was clearly visible because of a markedly enlarged pericerebral space associated with micrencephaly secondary to CMV infection. Bilateral opercular dysplasia was also present. Very few sonographic markers of infectious fetopathy were observed other than periventricular cysts located behind both ventricular horns. Magnetic resonance imaging (MRI) of the fetal brain confirmed the ultrasound findings and also showed the presence of marked micrencephaly, whereas cephalic measurements acquired on ultrasound examination (biparietal diameter and head circumference) were within the normal range. This case emphasizes the complementary roles of sonography and MRI in the prenatal diagnosis of cerebral abnormalities. Moreover, it illustrates the fact that polymicrogyria is easier to diagnose on ultrasound examination during the second trimester, before the development of secondary sulci.

Contribution of prenatal imaging to the anatomical assessment of fetal hydrocolpos

Hydrocolpos may be associated with a lower urinary tract obstruction in a spectrum of urorectal malformations ranging from persistent urogenital sinus to cloacal dysgenesis. As cloacal dysgenesis carries the worst postnatal prognosis, detailed prenatal ultrasound should focus on the fetal pelvic anatomy to provide the parents with appropriate prenatal counseling. We report three cases of fetal hydrocolpos associated with low urinary tract obstructions, including two with a normal appearance of the anal canal and rectum on prenatal ultrasound and one with a complex cloacal malformation which contributed to the precise prenatal assignment of the malformation in each case within the spectrum of urogenital sequence malformations.