Investigating Determinants and Evaluating Deep Learning Training Approaches for Visual Acuity in Foveal Hypoplasia

Perspectives on the Current and Future State of Artificial Intelligence in Medical Genetics

Artificial intelligence (AI) is rapidly transforming numerous aspects of daily life, including clinical practice and biomedical research. In light of this rapid transformation, and in the context of medical genetics, we assembled a group of leaders in the field to respond to the question about how AI is affecting, and especially how AI will affect, medical genetics. The authors who contributed to this collection of essays intentionally represent different areas of expertise, career stages, and geographies, and include diverse types of clinicians, computer scientists, and researchers. The individual pieces cover a wide range of areas related to medical genetics; we expect that these pieces may provide helpful windows into the ways in which AI is being actively studied, used, and considered in medical genetics.

Deep Learning Study of Alkaptonuria Spinal Disease Assesses Global and Regional Severity and Detects Occult Treatment Status

Deep learning (DL) is increasingly used to analyze medical imaging, but is less refined for rare conditions, which require novel pre-processing and analytical approaches. To assess DL in the context of rare diseases, this study focused on alkaptonuria (AKU), a rare disorder that affects the spine and involves other sequelae; treatments include the medication nitisinone. Since assessing x-rays to determine disease severity can be a slow, manual process requiring considerable expertise, this study aimed to determine whether these DL methods could accurately identify overall spine severity at specific regions of the spine and whether patients were receiving nitisinone. DL performance was evaluated versus clinical experts using cervical and lumbar spine radiographs. DL models predicted global severity scores (30-point scale) within 1.72 ± 1.96 points of expert clinician scores for cervical and 2.51 ± 1.96 points for lumbar radiographs. For region-specific metrics, the degrees of narrowing, calcium, and vacuum disc phenomena at each intervertebral space (IVS) were assessed. The model's narrowing scores were within 0.191-0.557 points from clinician scores (6-point scale), calcium was predicted with 78%-90% accuracy (present, absent, or disc fusion), and vacuum disc phenomenon predictions were less consistent (41%-90%). Intriguingly, DL models predicted nitisinone treatment status with 68%-77% accuracy, while expert clinicians appeared unable to discern nitisinone status (51% accuracy) (p = 2.0 × 10-9). This highlights the potential for DL to augment certain types of clinical assessments in rare disease, as well as identifying occult features like treatment status.

Generative AI for synthetic data across multiple medical modalities: A systematic review of recent developments and challenges

This paper presents a comprehensive systematic review of generative models (GANs, VAEs, DMs, and LLMs) used to synthesize various medical data types, including imaging (dermoscopic, mammographic, ultrasound, CT, MRI, and X-ray), text, time-series, and tabular data (EHR). Unlike previous narrowly focused reviews, our study encompasses a broad array of medical data modalities and explores various generative models. Our aim is to offer insights into their current and future applications in medical research, particularly in the context of synthesis applications, generation techniques, and evaluation methods, as well as providing a GitHub repository as a dynamic resource for ongoing collaboration and innovation. Our search strategy queries databases such as Scopus, PubMed, and ArXiv, focusing on recent works from January 2021 to November 2023, excluding reviews and perspectives. This period emphasizes recent advancements beyond GANs, which have been extensively covered in previous reviews. The survey also emphasizes the aspect of conditional generation, which is not focused on in similar work. Key contributions include a broad, multi-modality scope that identifies cross-modality insights and opportunities unavailable in single-modality surveys. While core generative techniques are transferable, we find that synthesis methods often lack sufficient integration of patient-specific context, clinical knowledge, and modality-specific requirements tailored to the unique characteristics of medical data. Conditional models leveraging textual conditioning and multimodal synthesis remain underexplored but offer promising directions for innovation. Our findings are structured around three themes: (1) Synthesis applications, highlighting clinically valid synthesis applications and significant gaps in using synthetic data beyond augmentation, such as for validation and evaluation; (2) Generation techniques, identifying gaps in personalization and cross-modality innovation; and (3) Evaluation methods, revealing the absence of standardized benchmarks, the need for large-scale validation, and the importance of privacy-aware, clinically relevant evaluation frameworks. These findings emphasize the need for benchmarking and comparative studies to promote openness and collaboration.

Foveal Hypoplasia Grading with Optical Coherence Tomography: Agreement and Challenges Across Experience Levels

Background/Objectives: The diagnosis and prognosis of arrested foveal development or foveal hypoplasia (FH) can be made using the Leicester grading system for FH and optical coherence tomography (OCT). In clinical practice, ophthalmologists and ophthalmic health professionals with varying experience consult patients with FH; however, to date, the FH grading system has only been validated amongst experts. We compare the inter-grader and intra-grade agreement of healthcare professionals against expert consensus across all grades of FH. Methods: Handheld and table-mounted OCT images (n = 341) were graded independently at a single centre by experts (n = 3) with over six years of experience and "novice" medical and allied health professionals (n = 5) with less than three years of experience. Sensitivity, specificity, and Cohen's kappa scores were calculated for each grader, and expert vs. novice performance was compared. Results: All graders showed high sensitivity (median 97% (IQR: 94-99)) and specificity (median 94% (IQR: 90-95)) in identifying the presence or absence of FH. No significant difference was seen in specificity between expert and novice graders, but experts had significantly greater diagnostic sensitivity (median difference = 5.3%, H = 5.00, p = 0.025). Expert graders had the highest agreement with the ground truth and novice graders showed great variability in grading uncommon grades, such as atypical FH. The proposed causes of misclassification included macular decentring in handheld OCT scans in children. Conclusions: Ophthalmologists of varying experience and allied health professionals can accurately identify FH using handheld and table-mounted OCT images. FH identification and paediatric OCT interpretation can be improved in wider ophthalmic clinical settings through the education of ophthalmic staff.

Deep Learning Study of Alkaptonuria Spinal Disease Assesses Global and Regional Severity and Detects Occult Treatment Status

Deep learning (DL) is increasingly used to analyze medical imaging, but is less refined for rare conditions, which require novel pre-processing and analytical approaches. To assess DL in the context of rare diseases, we focused on alkaptonuria (AKU), a rare disorder that affects the spine and involves other sequelae; treatments include the medication nitisinone. Since assessing X-rays to determine disease severity can be a slow, manual process requiring considerable expertise, we aimed to determine whether our DL methods could accurately identify overall spine severity, severity at specific regions of the spine, and whether DL could detect whether patients were receiving nitisinone. We evaluated DL performance versus clinical experts using cervical and lumbar spine radiographs. DL models predicted global severity scores (30-point scale) within 1.72 ± 1.96 points of expert clinician scores for cervical and 2.51 ± 1.96 points for lumbar radiographs. For region-specific metrics, we assessed the degree of narrowing, calcium, and vacuum phenomena at each intervertebral space (IVS). Our model's narrowing scores were within 0.191-0.557 points from clinician scores (6-point scale), calcium was predicted with 78-90% accuracy (present, absent, or disc fusion), while vacuum disc phenomenon predictions were less consistent (41-90%). Intriguingly, DL models predicted nitisinone treatment status with 68-77% accuracy, while expert clinicians appeared unable to discern nitisinone status (51% accuracy) (p = 2.0 × 10-9). This highlights the potential for DL to augment certain types of clinical assessments in rare disease, as well as identifying occult features like treatment status.

Artificial intelligence in clinical genetics

Artificial intelligence (AI) has been growing more powerful and accessible, and will increasingly impact many areas, including virtually all aspects of medicine and biomedical research. This review focuses on previous, current, and especially emerging applications of AI in clinical genetics. Topics covered include a brief explanation of different general categories of AI, including machine learning, deep learning, and generative AI. After introductory explanations and examples, the review discusses AI in clinical genetics in three main categories: clinical diagnostics; management and therapeutics; clinical support. The review concludes with short, medium, and long-term predictions about the ways that AI may affect the field of clinical genetics. Overall, while the precise speed at which AI will continue to change clinical genetics is unclear, as are the overall ramifications for patients, families, clinicians, researchers, and others, it is likely that AI will result in dramatic evolution in clinical genetics. It will be important for all those involved in clinical genetics to prepare accordingly in order to minimize the risks and maximize benefits related to the use of AI in the field.

Genetic Characteristics and Clinical Manifestations of Foveal Hypoplasia in Familial Exudative Vitreoretinopathy

PURPOSE

This study aimed to ascertain the occurrence of foveal hypoplasia (FH) in individuals diagnosed with familial exudative vitreoretinopathy (FEVR).

DESIGN

Retrospective cohort study.

METHODS

In this study, FEVR families and sporadic cases were diagnosed at the Eye and ENT Hospital, Fudan University, between 2017 and 2023. All patients attended routine ophthalmologic examinations and genetic screenings. The classification of FH was determined using optical coherence tomography (OCT) scans. The FH condition was classified into 2 subgroups: group A (FH being limited to the inner layers) and group B (FH affecting the outer layers). A total of 102 eyes from 58 patients were suitable for analysis.

RESULTS

Forty-nine mutations in LRP5, FZD4, NDP, TSPAN12, KIF11, CTNNB1, and ZNF408 were examined and detected, with 26 of them being novel. Forty-seven eyes (46.1%) revealed FH. The majority (53.2%) were due to the typical grade 1 FH. Patients with mutations in LRP5 and KIF11 were found to exhibit a higher prevalence of FH (P = .0088). Group B displayed the lowest visual acuity compared with group A (P = .048) and the group without FH (P < .001). The retinal arteriolar angle in group B was significantly smaller than in group A (P = .001) and those without FH (P < .001).

CONCLUSIONS

This study offers a new diagnostic approach and expands the spectrum of FEVR mutations. LRP5 and KIF11 were found to be more susceptible to causing FH in patients with FEVR. FEVR eyes with FH exhibited both greater visual impairment and reduced retinal arteriolar angles. The assessment of foveal status in patients with FEVR should be valued.

Quantitative Foveal Structural Metrics as Predictors of Visual Acuity in Human Albinism

Purpose

To assess the degree to which quantitative foveal structural measurements account for variation in best-corrected visual acuity (BCVA) in human albinism.

Methods

BCVA was measured and spectral domain optical coherence tomography (SD-OCT) images were acquired for 74 individuals with albinism. Categorical foveal hypoplasia grades were assessed using the Leicester Grading System for Foveal Hypoplasia. Foveal anatomical specialization (foveal versus parafoveal value) was quantified for inner retinal layer (IRL) thickness, outer segment (OS) length, and outer nuclear layer (ONL) thickness. These metrics, participant sex, and age were used to build a multiple linear regression of BCVA. This combined linear model's predictive properties were compared to those of categorical foveal hypoplasia grading.

Results

The cohort included three participants with type 1a foveal hypoplasia, 23 participants with type 1b, 33 with type 2, ten with type 3, and five with type 4. BCVA ranged from 0.08 to 1.00 logMAR (mean ± SD: 0.53 ± 0.21). IRL ratio, OS ratio, and ONL ratio were measured in all participants and decreased with increasing severity of foveal hypoplasia. The best-fit combined linear model included all three quantitative metrics and participant age expressed as a binary variable (divided into 0-18 years and 19 years or older; adjusted R2 = 0.500). This model predicted BCVA more accurately than a categorical foveal hypoplasia model (adjusted R2 = 0.352).

Conclusions

A quantitative model of foveal specialization accounts for more variance in BCVA in albinism than categorical foveal hypoplasia grading. Other factors, such as optical aberrations and eye movements, may account for the remaining unexplained variance.

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.

The future of commercial genetic testing

PURPOSE OF REVIEW

There are thousands of different clinical genetic tests currently available. Genetic testing and its applications continue to change rapidly for multiple reasons. These reasons include technological advances, accruing evidence about the impact and effects of testing, and many complex financial and regulatory factors.

RECENT FINDINGS

This article considers a number of key issues and axes related to the current and future state of clinical genetic testing, including targeted versus broad testing, simple/Mendelian versus polygenic and multifactorial testing models, genetic testing for individuals with high suspicion of genetic conditions versus ascertainment through population screening, the rise of artificial intelligence in multiple aspects of the genetic testing process, and how developments such as rapid genetic testing and the growing availability of new therapies for genetic conditions may affect the field.

SUMMARY

Genetic testing is expanding and evolving, including into new clinical applications. Developments in the field of genetics will likely result in genetic testing becoming increasingly in the purview of a very broad range of clinicians, including general paediatricians as well as paediatric subspecialists.

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.

SynthEye: Investigating the Impact of Synthetic Data on Artificial Intelligence-assisted Gene Diagnosis of Inherited Retinal Disease

Purpose

Rare disease diagnosis is challenging in medical image-based artificial intelligence due to a natural class imbalance in datasets, leading to biased prediction models. Inherited retinal diseases (IRDs) are a research domain that particularly faces this issue. This study investigates the applicability of synthetic data in improving artificial intelligence-enabled diagnosis of IRDs using generative adversarial networks (GANs).

Design

Diagnostic study of gene-labeled fundus autofluorescence (FAF) IRD images using deep learning.

Participants

Moorfields Eye Hospital (MEH) dataset of 15 692 FAF images obtained from 1800 patients with confirmed genetic diagnosis of 1 of 36 IRD genes.

Methods

A StyleGAN2 model is trained on the IRD dataset to generate 512 × 512 resolution images. Convolutional neural networks are trained for classification using different synthetically augmented datasets, including real IRD images plus 1800 and 3600 synthetic images, and a fully rebalanced dataset. We also perform an experiment with only synthetic data. All models are compared against a baseline convolutional neural network trained only on real data.

Main Outcome Measures

We evaluated synthetic data quality using a Visual Turing Test conducted with 4 ophthalmologists from MEH. Synthetic and real images were compared using feature space visualization, similarity analysis to detect memorized images, and Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) score for no-reference-based quality evaluation. Convolutional neural network diagnostic performance was determined on a held-out test set using the area under the receiver operating characteristic curve (AUROC) and Cohen's Kappa (κ).

Results

An average true recognition rate of 63% and fake recognition rate of 47% was obtained from the Visual Turing Test. Thus, a considerable proportion of the synthetic images were classified as real by clinical experts. Similarity analysis showed that the synthetic images were not copies of the real images, indicating that copied real images, meaning the GAN was able to generalize. However, BRISQUE score analysis indicated that synthetic images were of significantly lower quality overall than real images (P < 0.05). Comparing the rebalanced model (RB) with the baseline (R), no significant change in the average AUROC and κ was found (R-AUROC = 0.86[0.85-88], RB-AUROC = 0.88[0.86-0.89], R-k = 0.51[0.49-0.53], and RB-k = 0.52[0.50-0.54]). The synthetic data trained model (S) achieved similar performance as the baseline (S-AUROC = 0.86[0.85-87], S-k = 0.48[0.46-0.50]).

Conclusions

Synthetic generation of realistic IRD FAF images is feasible. Synthetic data augmentation does not deliver improvements in classification performance. However, synthetic data alone deliver a similar performance as real data, and hence may be useful as a proxy to real data. Financial Disclosure(s): Proprietary or commercial disclosure may be found after the references.