Phenotip - a web-based instrument to help diagnosing fetal syndromes antenatally

Prenatal ultrasound can often reliably distinguish fetal anatomic anomalies, particularly in the hands of an experienced ultrasonographer. Given the large number of existing syndromes and the significant overlap in prenatal findings, antenatal differentiation for syndrome diagnosis is difficult. We constructed a hierarchic tree of 1140 sonographic markers and submarkers, organized per organ system. Subsequently, a database of prenatally diagnosable syndromes was built. An internet-based search engine was then designed to search the syndrome database based on a single or multiple sonographic markers. Future developments will include a database with magnetic resonance imaging findings as well as further refinements in the search engine to allow prioritization based on incidence of syndromes and markers.


Background
Many countries have incorporated ultrasound in routine prenatal care for fetal anomaly screening. When multiple fetal anomalies are found, a syndrome is often suspected. Some syndromes have a known genetic background and can be identified by invasive fetal testing with routine karyotyping and/or comparative genomic hybridization (e.g. Edwards syndrome or DiGeorge syndrome). Many others however, require specific gene sequencing or do not have a known genetic origin (such as Noonan syndrome or Fryns syndrome) and cannot be identified by routine genetic screening tests. Accurate prenatal identification or suspicion of a syndrome is therefore important to guide further testing and/or counseling. Given the large number of known syndromes [1] (over 6000) and a significant overlap in prenatal findings, antenatal differentiation is difficult. The OMIM® (Online Mendelian Inheritance in Man) database [2], Orphanet [3], POSSUMweb [4] and London Medical Database [5] are searchable databases that allow links of phenotypic findings with (genetic) syndromes and may help in diagnosing syndromes. None of the database queries, however, include prenatal ultrasound findings (such as echogenic bowel or increased nuchal fold) in the search algorithm. Moreover, as these databases are mainly designed for postnatal use, they give great importance to markers that may not always be present or identifiable in the prenatal stage (such as failure to thrive, microcephaly or neurodevelopmental delay). Finally, these databases deal poorly with marker synonyms. As an example, the search terms "echogenic kidneys" and "hyperechogenic kidneys" yield 15 and 18 syndromes respectively in OMIM® [2], but only three syndromes are shared by both searches.
The need for a freely available tool, useable in the prenatal period, brought us to design 'Phenotip' , a free web-based searchable syndrome database, which is based exclusively on sonographic markers.

Database design
The Phenotip collaboration is an independent, international association between maternal-fetal medicine specialists with particular interest in prenatal diagnosis and a software engineer. The Phenotip database relies on a hierarchically structured "tree" of antenatal sonographic markers (n = 1140). Parent markers are organized by organ system and grow in resolution with every level of branching (daughter markers). For example, "face" branches into "eyes", "ears", "mouth and lips". "Mouth and lips" then further branches into "lip", "palate", "philtrum" and so on. Therefore, each marker has multiple parent and/or daughter markers. Marker synonyms have been defined to avoid confusion (e.g. talipesclubfoot). Overall, 1140 sonographic markers are available, among them 130 markers have at least one synonym. Each selected marker will appear on the left hand side of the screen under "selected marker" after clicking the green button. They can also be removed by clicking the red button. Differential diagnosis will appear after clicking the "show possible syndromes" button on the left hand side of the screen. Amniodrainage was performed, and CGH array confirmed a micro-deletion of locus p13.3 on chromosome 17 including LIS1gene.
Markers are grouped into syndromes based on an extensive literature search. Only markers that were previously described in a peer reviewed publication as part of the antenatal sonographic phenotype of a proven syndrome were included in the database. Each syndrome is defined by its specific daughter markers, but also includes all hierarchically superior parent markers.
When this information was available, we also noted the incidence and inheritance pattern and male/female ratio for each syndrome. Weblinks to relevant overview articles or websites such as OMIM® [2], Orphanet [3], Geneva Foundation [6], Jablonski's database [7] and SonoWorld [8] were added.
Information for each syndrome was registered by one editor, then peer-reviewed by at least one other editor. So far, we have collected literature on 329 of the most common syndromes.

Searching the database
The syndrome database is freely available through a web-based interface at www.phenotip.com. Users can search by syndrome name or by a combination of ultrasound markers.
When a specific marker is chosen, the search algorithm automatically includes all daughter markers of the chosen marker. Each level of the hierarchical tree of each specific organ system is thus considered. Choosing a parent marker will increase the sensitivity of the search while choosing a daughter marker will increase specificity. When a sonographic abnormality is not clearly defined, the involved organ can be selected, and hence all downstream markers would be considered. This is, for example, useful in cases of cardiac malformations, where one syndrome may present with a wide variety of heart lesions. Also, non-experienced sonographers might select the affected organ when they are unable to define the exact cardiac pathology.
Markers can either be selected from an expandable hierarchic tree or from a search box. Users can choose to search only syndromes including "all selected markers" or to search syndromes including either "one of the selected markers", thereby again increasing sensitivity or specificity, respectively.

Results
Since its inception in July 2013, the Phenotip database has logged 1215 sessions by 714 users with, among them, 136 regular visitors from 18 countries. The tool has allowed the identification of a sometimes-unsuspected diagnosis in many cases. A recent example suspected through our search algorithm and then confirmed by genetic analysis is presented in Figure 1.
Phenotip allows the search of differential diagnosis either by replacing a specific marker ("Flat nasal bridge") by a less specific marker ("Face"), or by removing a marker from the searching list. Table 1 provides an example based on markers used in Figure 1 (Miller-Dieker syndrome).
In addition, the database can offer guidance to the sonographer to find additional markers that differentiate between syndromes or genetic anomalies.
As this is a continuously evolving database (syndromes are being added on a daily basis), formal validation of sensitivity and specificity with validation against postnatal diagnosis has not been undertaken yet.
In order to test our database, all "cases of the week" from TheFetus.net were considered (380 cases, http:// sonoworld.com/TheFetus/Listing.aspx?Id=2). Inclusion Table 1 Differential diagnosis found using 4, 3 and 2 markers from Figure 1 Markers inserted in Phenotip.com Number of/diagnosis found with Phenotip.com (except Miller-Dieker Sd)

markers
Lissencephaly -Clinodactyly -Polyhydramnios -  criteria were syndromes only. Exclusion criteria were cases with a unique organ involved (mainly bone and heart). All the remaining cases were considered (n = 50, see Table 2). Only prenatal markers based on ultrasound images from the Fetus.net website were used in Phenotip. In 12 cases (24%), one of the markers used was only present in the prenatal period (such as hydramnios, single umbilical artery). Among 50 unselected cases, 43 (86%) were found as correct and unique diagnosis. In 7 cases (14%), 2-5 diagnoses were identified, always including the correct diagnosis. By using all the data provided in the Fetus.net (karyotype, recurrence), many of these multiple diagnoses can be excluded. We are currently adding new Phenotip functions such as "known karyotype" and "previous case in the family" to increase specificity (Table 2). Finally, the database is already designed to incorporate the relative frequency of each marker in each specific syndrome, so in the future the search will have even greater specificity and will use a Bayesian approach.

Conclusion
We here describe the development of a searchable database of fetal syndromes. In contrast to other (commercially) available databases, this database only relies on antenatally diagnosable markers and does not include often subtle, postnatal findings.
We feel that this database may help both more and less experienced sonographers, obstetricians, geneticists and fetal medicine specialists in reaching the diagnosis of a fetal syndrome antenatally. Indeed, medicine involves large amounts of data that usually have to be exploited jointly. Given the limitations of the human brain, complex mathematical algorithms or Bayesian networks [9], integrating all available information can obtain better diagnostic accuracy.
Computer assisted diagnosis is already put in clinical practice on a daily basis in other branches of obstetrics and gynecology. Examples of this include prenatal screening for trisomy 21 [10], outcome prediction of pregnancies of unknown location [11] and discriminating between benign and malignant ovarian masses [12].
This database certainly does not replace expert fetal care providers as it still requires the input of accurate findings and will often only generate a differential diagnosis, which then needs to be explored further. Moreover, dealing with computed knowledge and software as tools for diagnosis does not substitute communication skills and empathy when facing patients.
This project is a work in progress and the number of syndromes included in the database will be further updated. Future developments will include the addition of magnetic resonance imaging markers [13] as well as Trisomy 18, DiGeorge sd. (2) Cat-eye sd, Goldenhar sd, Branchio-oto-renal sd.
further refinements in the search engine to allow prioritisation based on incidence of syndromes and markers. Moreover, we will add postnatal findings and information to each syndrome.
We anticipate that the growing use of advanced technologies (such as chromosomal microarray [14] or exome sequencing [15]) for the prenatal diagnosis of genetic alterations that are associated with sonographic abnormalities will discover novel, currently unknown, syndromes. This will further enhance the linkage between specific sonographic findings and the concomitant genomic alteration. As data gathers, we will incorporate those novel syndromes and information into the database. With this database, we hope to facilitate antenatal diagnosis of fetal syndromes and improve patient care.

Presentation information
These data were presented at: