Presented by Omar Abdul-Rahman, MD, University of Nebraska Medical Center

Precision Medicine World Conference (PMWC)| September 24-25, 2018 | Duke University

This post is based on a presentation given at PMWC Duke. Watch the full presentation.

At the Precision Medicine World Conference (PMWC) Duke meeting, Dr. Omar Abdul-Rahman, Friedland Professor and Director of Genetic Medicine at the University of Nebraska Medical Center, told the story of the emergence of phenotypic data as crucial in clinical evaluations.

According to Dr. Abdul-Rahman, there are, “three legs to a stool that we have to understand” when making a patient evaluation: genes, environment, and phenotype. He described a time about ten years ago when “there had been a lot of advancements made in the ability to get good genomic data.” However, he went on to say that those same advancements had unfortunately not been made in phenotyping.

Standardizing the Phenotype

The first step taken in tackling this standstill and beginning to capture robust phenotypic information was the development of the Elements of Morphology as a way to “standardize the nomenclature” of phenotypes. Together, a group of geneticists defined over 400 features of the face, hands, and feet to streamline they way clinicians referred to various morphologies.

“Once we standardized the phenotype, the next question became, ‘How do we capture it?’”

Capturing the Phenotype

While at a study site for the National Children’s Study, Dr. Abdul-Rahman and his team realized that they were successfully capturing genetic and environmental data, but not phenotypic. With the goal of capturing as many of the 400+ features as possible, they showed 15 photos (eight of the head/neck, four of the hands, and three of the feet) and three videos to a panel of geneticists for review.

This laborious study was presented as a poster at a conference where it was strategically placed next to a poster focused on computer-aided facial recognition. Combining Dr. Abdul-Rahman understanding of how to capture enough imaging to analyze a phenotype with his colleagues’ computer-aided automation of the process, a collaborative study between the brains behind the neighboring posters was born.

Analyzing the Phenotype

The automating software introduced, Face2Gene is a suit of phenotyping applications developed by FDNA that operates on AI and deep learning technologies. Using facial analysis, the system has evaluated over 150,000 patients, and cross-comparing their phenotypes to a growing database of over 10,000 genetic diseases helps to enable more rapid and accurate diagnoses.

Studying FASD with Face2Gene

After learning about Face2Gene, Dr. Rahman chose to apply the technology to a particular teratogen of interest, alcohol, and related Fetal Alcohol Spectrum Disorders (FASD). FASD can be broken up into four diagnostic categories, the first three of which are straightforward:

• FAS – requires all of the typically-present features
• Partial FAS – requires some, but not all of the typically-present features
• Alcohol-related birth defects (ARBD) – includes a series of anomalies more common to present among alcohol consumers
• Alcohol-related neurodevelopmental disorder (ARND)

The final FASD category, Alcohol-related neurodevelopmental disorder (ARND), is more difficult to diagnose because there are no outward features. When diagnosing, there are two requirements: documented prenatal alcohol consumption and neurobehavior impairment, which cannot be diagnosed under the age of three.

Because ARND is both the most difficult FASD to diagnose and also the most prevalent FASD type in the United States, Dr. Abdul-Rahman decided to put Face2Gene to the test to see if the technology could detect the difference between the four diagnostic categories. He ran images of over 130 subjects with FAS, Partial FAS, ARBD, or ARND, as well as controls through Face2Gene in a cross-validation test. The images were split 50/50 into training and test sets and were run through ten rounds. The results showed that for any FASD vs. control, the manual and computer-aided score both performed well and relatively equally. The same was true of looking at the individual syndrome categories, FAS, Partial FAS, and ARBD; however, when looking at ARND vs. control, there was a better AUC for computer-aided vs. manual.

You can find the study here. 

Lessons Learned

Although there appeared to be no physical characteristics in the clinical criteria for ARND, Face2Gene was able to pick up on subtle facial cues, showing that, “there must be something that is broader, but is really subclinical from a provider standpoint.” Dr. Abdul-Rahman concluded with the following lessons learned from the study and his experience using Face2Gene:

• Facial analysis software like FDNA’s Face2Gene can be used to delineate phenotypes at a subclinical level
• Such software may allow for the identification of external physical biomarkers for conditions such as FASD
• Such software can potentially be used by primary care providers as a screen for further investigations (e.g. genetics referral)

Check out other presentations and similar research on the benefits of AI and facial analysis in personalized medicine.

The post Personalizing Medicine with Artificial Intelligence and Facial Analysis appeared first on FDNA™.

This post is based on a presentation given at the Precision Medicine World Conference. You can skip to the video by clicking here.

As researchers and clinicians expand their investigation of the relationship between phenotypes and genotypes, the importance of accurate phenotypic analysis grows. That means the field needs a uniform understanding of what phenotypes are, how to describe them, and how to assign traits.

In differentiating deep phenotyping and next-generation phenotyping, Dr. Donald Basel (Medical Director of the Genetics Center at the Children’s Hospital of Wisconsin) asked Precision Medicine World Conference attendees, “Where do we limit our thoughts as to what a phenotype represents? Is it limited to structural morphology or do we consider all aspects of the phenotype?”

Deep phenotyping, which has existed since the 1950s, made a leap forward in dysmorphology when in 2009 NHGRI developed a common language, the hierarchical Human Phenotype Ontology (HPO). Next-generation phenotyping expands that concept, including not just structural morphology but aspects like speech analysis, gait analysis, and all of the “omics”–metabolomics, microbiomics, etc.

Even with the benefits of a common language, clinicians are still faced with the challenge of properly applying the HPO terms. Dr. Basel (also a member of the FDNA Scientific Advisory Board) noted that a tool like Face2Gene, a suite of phenotyping applications, does some of this identification for the clinician, and gave an example of human vs machine analysis of Cornelia de Lange syndrome and phenotypically similar syndromes. A panel of experts identified patients with Cornelia de Lange 77 percent of the time and spotted similar syndromes 87 percent of the time, with a clinical sensitivity of 82 percent and a specificity of 89 percent, whereas FDNA’s DeepGestalt technology correctly selected Cornelia de Lange 94 percent of the time and similar phenotypes 100 percent of the time, with a sensitivity of 86 percent and a specificity of 100 percent.

Of course, as Dr. Basel pointed out when it comes to Cornelia de Lange, a well-trained dysmorphologist can “essentially walk into a baby’s room and make this diagnosis.” The more profound results are for patients with phenotypes that would be far more difficult to diagnose because of the subtlety of facial traits, or for helping non-dysmorphologists recognize these phenotypes.

In a March 2018 publication, Dr. Basel described as “quite daring,” researchers took four syndromes with inborn metabolic errors and mild facial feature coarsening, plus Nicolaides-Baraitser syndrome (which has a similar phenotype) to test the

“Just looking at pure facial imaging the software was able to accurately predict the specific disorder in 64 percent of the cases. If you added a single feature into that feature set, the diagnostic accuracy increased to 87 percent.”

This means clinicians using Face2Gene may be presented with syndrome recommendations that they might have otherwise disregarded.

Similarly to how clinicians gain expertise with practice, Face2Gene grows smarter as it sees more cases. “The more we use it, the better it gets,” Basel said.

This brought him back to his original question: if artificial intelligence technologies can grow to analyze more and more data of increasingly varied types, “Where are the limits?”

Donald Basel presents at PMWC

The post “Where are the limits?” appeared first on FDNA™.