Approximating facial expression effects on diagnostic accuracy via generative AI in medical genetics.

Artificial intelligence (AI) is increasingly used in genomics research and practice, and generative AI has garnered significant recent attention. In clinical applications of generative AI, aspects of the underlying datasets can impact results, and confounders should be studied and mitigated. One example involves the facial expressions of people with genetic conditions. Stereotypically, Williams (WS) and Angelman (AS) syndromes are associated with a "happy" demeanor, including a smiling expression. Clinical geneticists may be more likely to identify these conditions in images of smiling individuals. To study the impact of facial expression, we analyzed publicly available facial images of approximately 3500 individuals with genetic conditions. Using a deep learning (DL) image classifier, we found that WS and AS images with non-smiling expressions had significantly lower prediction probabilities for the correct syndrome labels than those with smiling expressions. This was not seen for 22q11.2 deletion and Noonan syndromes, which are not associated with a smiling expression. To further explore the effect of facial expressions, we computationally altered the facial expressions for these images. We trained HyperStyle, a GAN-inversion technique compatible with StyleGAN2, to determine the vector representations of our images. Then, following the concept of InterfaceGAN, we edited these vectors to recreate the original images in a phenotypically accurate way but with a different facial expression. Through online surveys and an eye-tracking experiment, we examined how altered facial expressions affect the performance of human experts. We overall found that facial expression is associated with diagnostic accuracy variably in different genetic conditions.

Recognition of Genetic Conditions After Learning With Images Created Using Generative Artificial Intelligence.

Importance: The lack of standardized genetics training in pediatrics residencies, along with a shortage of medical geneticists, necessitates innovative educational approaches. Objective: To compare pediatric resident recognition of Kabuki syndrome (KS) and Noonan syndrome (NS) after 1 of 4 educational interventions, including generative artificial intelligence (AI) methods. Design, Setting, and Participants: This comparative effectiveness study used generative AI to create images of children with KS and NS. From October 1, 2022, to February 28, 2023, US pediatric residents were provided images through a web-based survey to assess whether these images helped them recognize genetic conditions. Interventions: Participants categorized 20 images after exposure to 1 of 4 educational interventions (text-only descriptions, real images, and 2 types of images created by generative AI). Main Outcomes and Measures: Associations between educational interventions with accuracy and self-reported confidence. Results: Of 2515 contacted pediatric residents, 106 and 102 completed the KS and NS surveys, respectively. For KS, the sensitivity of text description was 48.5% (128 of 264), which was not significantly different from random guessing (odds ratio [OR], 0.94; 95% CI, 0.69-1.29; P = .71). Sensitivity was thus compared for real images vs random guessing (60.3% [188 of 312]; OR, 1.52; 95% CI, 1.15-2.00; P = .003) and 2 types of generative AI images vs random guessing (57.0% [212 of 372]; OR, 1.32; 95% CI, 1.04-1.69; P = .02 and 59.6% [193 of 324]; OR, 1.47; 95% CI, 1.12-1.94; P = .006) (denominators differ according to survey responses). The sensitivity of the NS text-only description was 65.3% (196 of 300). Compared with text-only, the sensitivity of the real images was 74.3% (205 of 276; OR, 1.53; 95% CI, 1.08-2.18; P = .02), and the sensitivity of the 2 types of images created by generative AI was 68.0% (204 of 300; OR, 1.13; 95% CI, 0.77-1.66; P = .54) and 71.0% (247 of 328; OR, 1.30; 95% CI, 0.92-1.83; P = .14). For specificity, no intervention was statistically different from text only. After the interventions, the number of participants who reported being unsure about important diagnostic facial features decreased from 56 (52.8%) to 5 (7.6%) for KS (P < .001) and 25 (24.5%) to 4 (4.7%) for NS (P < .001). There was a significant association between confidence level and sensitivity for real and generated images. Conclusions and Relevance: In this study, real and generated images helped participants recognize KS and NS; real images appeared most helpful. Generated images were noninferior to real images and could serve an adjunctive role, particularly for rare conditions.

Comparison of clinical geneticist and computer visual attention in assessing genetic conditions.

Artificial intelligence (AI) for facial diagnostics is increasingly used in the genetics clinic to evaluate patients with potential genetic conditions. Current approaches focus on one type of AI called Deep Learning (DL). While DL- based facial diagnostic platforms have a high accuracy rate for many conditions, less is understood about how this technology assesses and classifies (categorizes) images, and how this compares to humans. To compare human and computer attention, we performed eye-tracking analyses of geneticist clinicians (n = 22) and non-clinicians (n = 22) who viewed images of people with 10 different genetic conditions, as well as images of unaffected individuals. We calculated the Intersection-over-Union (IoU) and Kullback-Leibler divergence (KL) to compare the visual attentions of the two participant groups, and then the clinician group against the saliency maps of our deep learning classifier. We found that human visual attention differs greatly from DL model's saliency results. Averaging over all the test images, IoU and KL metric for the successful (accurate) clinician visual attentions versus the saliency maps were 0.15 and 11.15, respectively. Individuals also tend to have a specific pattern of image inspection, and clinicians demonstrate different visual attention patterns than non-clinicians (IoU and KL of clinicians versus non-clinicians were 0.47 and 2.73, respectively). This study shows that humans (at different levels of expertise) and a computer vision model examine images differently. Understanding these differences can improve the design and use of AI tools, and lead to more meaningful interactions between clinicians and AI technologies.

Generative Methods for Pediatric Genetics Education.

Artificial intelligence (AI) is used in an increasing number of areas, with recent interest in generative AI, such as using ChatGPT to generate programming code or DALL-E to make illustrations. We describe the use of generative AI in medical education. Specifically, we sought to determine whether generative AI could help train pediatric residents to better recognize genetic conditions. From publicly available images of individuals with genetic conditions, we used generative AI methods to create new images, which were checked for accuracy with an external classifier. We selected two conditions for study, Kabuki (KS) and Noonan (NS) syndromes, which are clinically important conditions that pediatricians may encounter. In this study, pediatric residents completed 208 surveys, where they each classified 20 images following exposure to one of 4 possible educational interventions, including with and without generative AI methods. Overall, we find that generative images perform similarly but appear to be slightly less helpful than real images. Most participants reported that images were useful, although real images were felt to be more helpful. We conclude that generative AI images may serve as an adjunctive educational tool, particularly for less familiar conditions, such as KS.

Application of facial analysis Technology in Clinical Genetics: Considerations for diverse populations.

Facial analysis technology in rare diseases has the potential to shorten the diagnostic odyssey by providing physicians with a valuable diagnostic tool. Given that most clinical genetic resources focus on populations of European descent, we compare craniofacial features in genetic syndromes across different populations and review how machine learning algorithms perform on diagnosing genetic syndromes in geographically and ethnically diverse populations. We also discuss the value of populations from ancestrally diverse backgrounds in the training set of machine learning algorithms. Finally, this review demonstrates that across diverse population groups, machine learning models have outstanding accuracy as supported by the area under the curve values greater than 0.9. Artificial intelligence is only in its infancy in the diagnosis of rare disease in diverse populations and will become more accurate as larger and more diverse training sets, including a wider spectrum of ages, particularly infants, are studied.

Human and computer attention in assessing genetic conditions.

Deep learning (DL) and other types of artificial intelligence (AI) are increasingly used in many biomedical areas, including genetics. One frequent use in medical genetics involves evaluating images of people with potential genetic conditions to help with diagnosis. A central question involves better understanding how AI classifiers assess images compared to humans. To explore this, we performed eye-tracking analyses of geneticist clinicians and non-clinicians. We compared results to DL-based saliency maps. We found that human visual attention when assessing images differs greatly from the parts of images weighted by the DL model. Further, individuals tend to have a specific pattern of image inspection, and clinicians demonstrate different visual attention patterns than non-clinicians.

Perspectives on the future of dysmorphology.

The field of clinical genetics and genomics continues to evolve. In the past few decades, milestones like the initial sequencing of the human genome, dramatic changes in sequencing technologies, and the introduction of artificial intelligence, have upended the field and offered fascinating new insights. Though difficult to predict the precise paths the field will follow, rapid change may continue to be inevitable. Within genetics, the practice of dysmorphology, as defined by pioneering geneticist David W. Smith in the 1960s as "the study of, or general subject of abnormal development of tissue form" has also been affected by technological advances as well as more general trends in biomedicine. To address possibilities, potential, and perils regarding the future of dysmorphology, a group of clinical geneticists, representing different career stages, areas of focus, and geographic regions, have contributed to this piece by providing insights about how the practice of dysmorphology will develop over the next several decades.

Bridging the language gap - A call for the wider use of Human phenotype ontology by non-geneticist clinicians when requesting genomic tests.

Advances in genomic technology including the development of next-generation sequencing (NGS) have enabled the identification of thousands of variations at a time, allowing the discovery of novel genetic diseases. Given the volume of data generated by these investigations, attention is drawn towards reporting relevant clinical features by clinicians to guide the diagnosis and management of their patients. The Human Phenotype Ontology (HPO) developed in 2008, revolutionized the semantic vocabulary of phenotypic descriptions in genomic medicine allowing researchers, laboratories and clinical geneticists to better understand each other. In this era of personalized medicine where genetic tests are becoming more accessible, non-geneticist clinicians are expected to be more involved than ever in the process of ordering genetic tests and interpreting genetic reports. It is therefore essential that they understand and adequately apply HPO nomenclature to integrate the patient care chain and seize the opportunity offered by this tailored language. The current article highlights the importance of using HPO vocabularies in clinical practice and advocates for its wider use by non-geneticist clinicians. Correct use of HPO will reduce misunderstandings between healthcare professionals and ultimately improve the healthcare system.

Scoping review and classification of deep learning in medical genetics.

Deep learning (DL) is applied in many biomedical areas. We performed a scoping review on DL in medical genetics. We first assessed 14,002 articles, of which 133 involved DL in medical genetics. DL in medical genetics increased rapidly during the studied period. In medical genetics, DL has largely been applied to small data sets of affected individuals (mean = 95, median = 29) with genetic conditions (71 different genetic conditions were studied; 24 articles studied multiple conditions). A variety of data types have been used in medical genetics, including radiologic (20%), ophthalmologic (14%), microscopy (8%), and text-based data (4%); the most common data type was patient facial photographs (46%). DL authors and research subjects overrepresent certain geographic areas (United States, Asia, and Europe). Convolutional neural networks (89%) were the most common method. Results were compared with human performance in 31% of studies. In total, 51% of articles provided data access; 16% released source code. To further explore DL in genomics, we conducted an additional analysis, the results of which highlight future opportunities for DL in medical genetics. Finally, we expect DL applications to increase in the future. To aid data curation, we evaluated a DL, random forest, and rule-based classifier at categorizing article abstracts.

Neural Networks for Classification and Image Generation of Aging in Genetic Syndromes.

Background: In medical genetics, one application of neural networks is the diagnosis of genetic diseases based on images of patient faces. While these applications have been validated in the literature with primarily pediatric subjects, it is not known whether these applications can accurately diagnose patients across a lifespan. We aimed to extend previous works to determine whether age plays a factor in facial diagnosis as well as to explore other factors that may contribute to the overall diagnostic accuracy. Methods: To investigate this, we chose two relatively common conditions, Williams syndrome and 22q11.2 deletion syndrome. We built a neural network classifier trained on images of affected and unaffected individuals of different ages and compared classifier accuracy to clinical geneticists. We analyzed the results of saliency maps and the use of generative adversarial networks to boost accuracy. Results: Our classifier outperformed clinical geneticists at recognizing face images of these two conditions within each of the age groups (the performance varied between the age groups): 1) under 2 years old, 2) 2-9 years old, 3) 10-19 years old, 4) 20-34 years old, and 5) ≥35 years old. The overall accuracy improvement by our classifier over the clinical geneticists was 15.5 and 22.7% for Williams syndrome and 22q11.2 deletion syndrome, respectively. Additionally, comparison of saliency maps revealed that key facial features learned by the neural network differed with respect to age. Finally, joint training real images with multiple different types of fake images created by a generative adversarial network showed up to 3.25% accuracy gain in classification accuracy. Conclusion: The ability of clinical geneticists to diagnose these conditions is influenced by the age of the patient. Deep learning technologies such as our classifier can more accurately identify patients across the lifespan based on facial features. Saliency maps of computer vision reveal that the syndromic facial feature attributes change with the age of the patient. Modest improvements in the classifier accuracy were observed when joint training was carried out with both real and fake images. Our findings highlight the need for a greater focus on age as a confounder in facial diagnosis.

Neural network classifiers for images of genetic conditions with cutaneous manifestations.

Neural networks have shown strong potential in research and in healthcare. Mainly due to the need for large datasets, these applications have focused on common medical conditions, where more data are typically available. Leveraging publicly available data, we trained a neural network classifier on images of rare genetic conditions with skin findings. We used approximately 100 images per condition to classify 6 different genetic conditions. We analyzed both preprocessed images that were cropped to show only the skin lesions as well as more complex images showing features such as the entire body segment, the person, and/or the background. The classifier construction process included attribution methods to visualize which pixels were most important for computer-based classification. Our classifier was significantly more accurate than pediatricians or medical geneticists for both types of images and suggests steps for further research involving clinical scenarios and other applications.

Beyond Trinucleotide Repeat Expansion in Fragile X Syndrome: Rare Coding and Noncoding Variants in FMR1 and Associated Phenotypes.

FMR1 (FMRP translational regulator 1) variants other than repeat expansion are known to cause disease phenotypes but can be overlooked if they are not accounted for in genetic testing strategies. We collected and reanalyzed the evidence for pathogenicity of FMR1 coding, noncoding, and copy number variants published to date. There is a spectrum of disease-causing FMR1 variation, with clinical and functional evidence supporting pathogenicity of five splicing, five missense, one in-frame deletion, one nonsense, and four frameshift variants. In addition, FMR1 deletions occur in both mosaic full mutation patients and as constitutional pathogenic alleles. De novo deletions arise not only from full mutation alleles but also alleles with normal-sized CGG repeats in several patients, suggesting that the CGG repeat region may be prone to genomic instability even in the absence of repeat expansion. We conclude that clinical tests for potentially FMR1-related indications such as intellectual disability should include methods capable of detecting small coding, noncoding, and copy number variants.

A Monoallelic Variant in REST Is Associated with Non-Syndromic Autosomal Dominant Hearing Impairment in a South African Family.

Hearing impairment (HI) is a sensory disorder with a prevalence of 0.0055 live births in South Africa. DNA samples from a South African family presenting with progressive, autosomal dominant non-syndromic HI were subjected to whole-exome sequencing, and a novel monoallelic variant in REST [c.1244GC; p.(C415S)], was identified as the putative causative variant. The co-segregation of the variant was confirmed with Sanger Sequencing. The variant is absent from databases, 103 healthy South African controls, and 52 South African probands with isolated HI. In silico analysis indicates that the p.C415S variant in REST substitutes a conserved cysteine and results in changes to the surrounding secondary structure and the disulphide bonds, culminating in alteration of the tertiary structure of REST. Localization studies using ectopically expressed GFP-tagged Wild type (WT) and mutant REST in HEK-293 cells show that WT REST localizes exclusively to the nucleus; however, the mutant protein localizes throughout the cell. Additionally, mutant REST has an impaired ability to repress its known target AF1q. The data demonstrates that the identified mutation compromises the function of REST and support its implication in HI. This study is the second report, worldwide, to implicate REST in HI and suggests that it should be included in diagnostic HI panels.

Cross-cultural representations of conjoined twins.

Conjoined twinning is a rare birth defect estimated to occur in about 1 in 50,000 to 100,000 births. The mechanism of conjoined twinning is not proven. Different forms of conjoined twinning are observed with the thoracopagus form being the most common. The rate of conjoined twinning is similar across all major populations. A dramatic malformation of this type would be an extraordinary occurrence leading people to reflect on the spiritual or supernatural nature of such an event. Therefore, it is not surprising that artifacts that seem to depict different forms of conjoined twins are seen across diverse cultures. In this article, we present a survey of these cultural artifacts including anatomic classification based on external anatomy and an exploration of the cultural and spiritual contexts associated with the artifacts. A key finding is that the most common form of conjoined twinning in the artifacts is parapagus (both dicephalus and diprosopus) in contrast to thoracopagus, the most common form in epidemiologic studies. Potential reasons for this difference are discussed. Evidence is presented to support the speculation that these objects represent artistic impressions of actual conjoined twinning events.

Exome Sequencing and Congenital Heart Disease in Sub-Saharan Africa.

BACKGROUND: Congenital heart disease (CHD) is the most common birth defect and affects roughly 1% of the global population. There have been many large CHD sequencing projects in developing countries but none in sub-Saharan Africa. In this exome sequencing study, we recruited families from Lagos, Nigeria, affected by structural heart disease. METHODS: Ninety-eight participants with CHD and an average age of 3.6 years were recruited from Lagos, Nigeria. Exome sequencing was performed on probands and parents when available. For genes of high interest, we conducted functional studies in Drosophila using a cardiac-specific RNA interference-based gene silencing system. RESULTS: The 3 most common CHDs were tetralogy of Fallot (20%), isolated ventricular septal defect (14%), and transposition of the great arteries (8%). Ten percent of the cohort had pathogenic or likely pathogenic variants in genes known to cause CHD. In 64 complete trios, we found 34 de novo variants that were not present in the African population in the Genome Aggregation Database (v3). Nineteen loss of function variants were identified using the genome-wide distribution of selection effects for heterozygous protein-truncating variants (shet). Nine genes caused a significant mortality when silenced in the Drosophila heart, including 4 novel disease genes not previously associated with CHD (UBB, EIF4G3, SREBF1, and METTL23). CONCLUSIONS: This study identifies novel candidate genes and variants for CHD and facilitates comparisons with previous CHD sequencing studies in predominantly European cohorts. The study represents an important first step in genomic studies of CHD in understudied populations. Registration: URL: https://www.clinicaltrials.gov; Unique identifier: NCT01952171.

Rubinstein-Taybi syndrome in diverse populations.

Rubinstein-Taybi syndrome (RSTS) is an autosomal dominant disorder, caused by loss-of-function variants in CREBBP or EP300. Affected individuals present with distinctive craniofacial features, broad thumbs and/or halluces, and intellectual disability. RSTS phenotype has been well characterized in individuals of European descent but not in other populations. In this study, individuals from diverse populations with RSTS were assessed by clinical examination and facial analysis technology. Clinical data of 38 individuals from 14 different countries were analyzed. The median age was 7 years (age range: 7 months to 47 years), and 63% were females. The most common phenotypic features in all population groups included broad thumbs and/or halluces in 97%, convex nasal ridge in 94%, and arched eyebrows in 92%. Face images of 87 individuals with RSTS (age range: 2 months to 47 years) were collected for evaluation using facial analysis technology. We compared images from 82 individuals with RSTS against 82 age- and sex-matched controls and obtained an area under the receiver operating characteristic curve (AUC) of 0.99 (p < .001), demonstrating excellent discrimination efficacy. The discrimination was, however, poor in the African group (AUC: 0.79; p = .145). Individuals with EP300 variants were more effectively discriminated (AUC: 0.95) compared with those with CREBBP variants (AUC: 0.93). This study shows that clinical examination combined with facial analysis technology may enable earlier and improved diagnosis of RSTS in diverse populations.

Noonan syndrome on the African Continent.

BACKGROUND: Noonan syndrome is a common genetic syndrome caused by pathogenic variants in genes in the Ras/MAPK signaling pathway. The medical literature has an abundance of studies on Noonan syndrome, but few are from the African continent. METHODS: The medical literature was searched for studies on Noonan syndrome from the African continent and these reports were added to our experience in Africa. Facial analysis was reviewed from two previous reports from our group using a support vector machine (SVM) algorithm and an analysis using the Face2Gene convolutional neural network technology. RESULTS: Individuals with Noonan syndrome from reports in African populations have the classic phenotype characteristics including typical minor facial anomalies such as widely spaced eyes (31-100%), short stature (71-100%), and congenital heart disease with pulmonary stenosis found in 24-100% of patients. Similarly, the genotypes are similar with the majority of variants occurring in the gene PTPN11 (72%) and 36% of these variants occurred in the amino acid residue Asn308, which is most commonly found in other populations. The two separate facial analysis algorithms successfully discriminated Africans with NS from unaffected matched individuals with area under the curve (AUC) of the receiver operator characteristic of 0.94 (SVM) and 0.979 for the Face2Gene research methodology. CONCLUSION: Few studies characterizing Noonan syndrome in Africans have been conducted, highlighting the need for more genetic and genomic research in African populations. Available clinical data, genotypes, and facial analysis technology data show that individuals of African descent with NS can be efficiently diagnosed using available standards.

Comorbidity of congenital heart defects and holoprosencephaly is likely genetically driven and gene-specific.

Comorbidity of holoprosencephaly (HPE) and congenital heart disease (CHD) in individuals with genetic variants in known HPE-related genes has been recurrently observed. Morphogenesis of the brain and heart from very early stages are regulated by several biological pathways, some of them involved in both heart and brain development as evidenced by genetic studies on model organisms. For instance, downregulation of Hedgehog or Nodal signaling pathways, both known as major triggers of HPE, has been shown to play a role in the pathogenesis of CHD, including structural defects and left-right asymmetry defects. In this study, individuals with various types of HPE were investigated clinically and by genomic sequencing. Cardiac phenotypes were assessed in 434 individuals with HPE who underwent targeted sequencing. CHDs were identified in 8% (n = 33) of individuals, including 10 (30%) cases of complex heart disease. Only four individuals (4/33) had damaging variants in the known HPE genes STAG2, SIX3, and SHH. Interestingly, no CHD was identified in the 37 individuals of our cohort with pathogenic variants in ZIC2. These findings suggest that CHD occurs more frequently in HPE-affected individuals with or without identifiable genetic variants, and this co-occurrence may be genetically driven and gene-specific.