Explore the vast range of titles available.

Deep learning may transform health care, but model development has largely been dependent on availability of advanced technical expertise. Herein we present the development of a deep learning model by clinicians without coding, which predicts reported sex from retinal fundus photographs. A model was trained on 84,743 retinal fundus photos from the UK Biobank dataset. External validation was performed on 252 fundus photos from a tertiary ophthalmic referral center. For internal validation, the area under the receiver operating characteristic curve (AUROC) of the code free deep learning (CFDL) model was 0.93. Sensitivity, specificity, positive predictive value (PPV) and accuracy (ACC) were 88.8%, 83.6%, 87.3% and 86.5%, and for external validation were 83.9%, 72.2%, 78.2% and 78.6% respectively. Clinicians are currently unaware of distinct retinal feature variations between males and females, highlighting the importance of model explainability for this task. The model performed significantly worse when foveal pathology was present in the external validation dataset, ACC: 69.4%, compared to 85.4% in healthy eyes, suggesting the fovea is a salient region for model performance OR (95% CI): 0.36 (0.19, 0.70) p = 0.0022. Automated machine learning (AutoML) may enable clinician-driven automated discovery of novel insights and disease biomarkers.

Retinal diseases are a leading cause of blindness in developed countries, accounting for the largest share of visually impaired children, working-age adults (inherited retinal disease), and elderly individuals (age-related macular degeneration). These conditions need specialised clinicians to interpret multimodal retinal imaging, with diagnosis and intervention potentially delayed. With an increasing and ageing population, this is becoming a global health priority. One solution is the development of artificial intelligence (AI) software to facilitate rapid data processing. Herein, we review research offering decision support for the diagnosis, classification, monitoring, and treatment of retinal disease using AI. We have prioritised diabetic retinopathy, age-related macular degeneration, inherited retinal disease, and retinopathy of prematurity. There is cautious optimism that these algorithms will be integrated into routine clinical practice to facilitate access to vision-saving treatments, improve efficiency of healthcare systems, and assist clinicians in processing the ever-increasing volume of multimodal data, thereby also liberating time for doctor-patient interaction and co-development of personalised management plans.

A number of large technology companies have created code-free cloud-based platforms that allow researchers and clinicians without coding experience to create deep learning algorithms. In this study, we comprehensively analyse the performance and featureset of six platforms, using four representative cross-sectional and en-face medical imaging datasets to create image classification models. The mean (s.d.) F1 scores across platforms for all model–dataset pairs were as follows: Amazon, 93.9 (5.4); Apple, 72.0 (13.6); Clarifai, 74.2 (7.1); Google, 92.0 (5.4); MedicMind, 90.7 (9.6); Microsoft, 88.6 (5.3). The platforms demonstrated uniformly higher classification performance with the optical coherence tomography modality. Potential use cases given proper validation include research dataset curation, mobile ‘edge models’ for regions without internet access, and baseline models against which to compare and iterate bespoke deep learning approaches. Several technology companies offer platforms for users without coding experience to develop deep learning algorithms. This Analysis compares the performance of six ‘code-free deep learning’ platforms (from Amazon, Apple, Clarifai, Google, MedicMind and Microsoft) in creating medical image classification models.

Background/AimsTo investigate the utility of a data-driven deep learning approach in patients with inherited retinal disorder (IRD) and to predict the causative genes based on fundus photography and fundus autofluorescence (FAF) imaging.MethodsClinical and genetic data from 1302 subjects from 729 genetically confirmed families with IRD registered with the Japan Eye Genetics Consortium were reviewed. Three categories of genetic diagnosis were selected, based on the high prevalence of their causative genes: Stargardt disease (ABCA4), retinitis pigmentosa (EYS) and occult macular dystrophy (RP1L1). Fundus photographs and FAF images were cropped in a standardised manner with a macro algorithm. Images for training/testing were selected using a randomised, fourfold cross-validation method. The application program interface was established to reach the learning accuracy of concordance (target: >80%) between the genetic diagnosis and the machine diagnosis (ABCA4, EYS, RP1L1 and normal).ResultsA total of 417 images from 156 Japanese subjects were examined, including 115 genetically confirmed patients caused by the three prevalent causative genes and 41 normal subjects. The mean overall test accuracy for fundus photographs and FAF images was 88.2% and 81.3%, respectively. The mean overall sensitivity/specificity values for fundus photographs and FAF images were 88.3%/97.4% and 81.8%/95.5%, respectively.ConclusionA novel application of deep neural networks in the prediction of the causative IRD genes from fundus photographs and FAF, with a high prediction accuracy of over 80%, was highlighted. These achievements will extensively promote the quality of medical care by facilitating early diagnosis, especially by non-specialists, access to care, reducing the cost of referrals, and preventing unnecessary clinical and genetic testing.

ObjectiveTo assess the capability of a convolutional neural network trained by transfer learning on anterior segment optical coherence tomography (AS-OCT) images, Placido-disk corneal topography images and external photographs to predict age and biological sex.DesignDevelopment of a deep learning model trained on retrospectively collected data using transfer learning.SettingA multicentre secondary care public health trust based in London.ParticipantsWe included 557,468 scans from 40,592 eyes of 20,542 patients. Data were extracted from all patients who underwent MS-39 imaging within our trust from October 2020 to March 2023.Primary and secondary outcome measuresPrimary outcome measures for biological sex classification included accuracy, precision, recall, F1-score and area under the receiver operating curve (ROC-AUC). Primary outcome measures for age prediction were Pearson correlation coefficients (r), coefficients of determination (R²) and the mean absolute error (MAE) to evaluate the predictive performance. The secondary outcome was to visualise and interpret the model’s decision-making process through the construction of saliency maps.ResultsFor age prediction, the MAEs for the Placido, AS-OCT and external photograph models were 5.2, 5.1 and 6.2 years, respectively. For gender classification, the same models achieved ROC-AUCs of 0.88, 0.73 and 0.81, respectively. No difference in performance was found in the analysis of corneas with pathological topography. The saliency maps highlighted the peri-limbal cornea for age prediction and the central cornea for gender discrimination.ConclusionsOur study demonstrates that deep learning models can extract age and gender information from anterior segment images. These findings support the concept that the anterior segment, like the retina, encodes important biological information. Future research should explore whether these models can predict specific systemic conditions.

The use of structured electronic health records in health care systems has grown rapidly. These systems collect huge amounts of patient information, including diagnosis codes representing temporal medical history. Sequential diagnostic information has proven valuable for predicting patient outcomes. However, the extent to which these types of data have been incorporated into deep learning (DL) models has not been examined. This systematic review aims to describe the use of sequential diagnostic data in DL models, specifically to understand how these data are integrated, whether sample size improves performance, and whether the identified models are generalizable. Relevant studies published up to May 15, 2023, were identified using 4 databases: PubMed, Embase, IEEE Xplore, and Web of Science. We included all studies using DL algorithms trained on sequential diagnosis codes to predict patient outcomes. We excluded review articles and non-peer-reviewed papers. We evaluated the following aspects in the included papers: DL techniques, characteristics of the dataset, prediction tasks, performance evaluation, generalizability, and explainability. We also assessed the risk of bias and applicability of the studies using the Prediction Model Study Risk of Bias Assessment Tool (PROBAST). We used the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist to report our findings. Of the 740 identified papers, 84 (11.4%) met the eligibility criteria. Publications in this area increased yearly. Recurrent neural networks (and their derivatives; 47/84, 56%) and transformers (22/84, 26%) were the most commonly used architectures in DL-based models. Most studies (45/84, 54%) presented their input features as sequences of visit embeddings. Medications (38/84, 45%) were the most common additional feature. Of the 128 predictive outcome tasks, the most frequent was next-visit diagnosis (n=30, 23%), followed by heart failure (n=18, 14%) and mortality (n=17, 13%). Only 7 (8%) of the 84 studies evaluated their models in terms of generalizability. A positive correlation was observed between training sample size and model performance (area under the receiver operating characteristic curve; P=.02). However, 59 (70%) of the 84 studies had a high risk of bias. The application of DL for advanced modeling of sequential medical codes has demonstrated remarkable promise in predicting patient outcomes. The main limitation of this study was the heterogeneity of methods and outcomes. However, our analysis found that using multiple types of features, integrating time intervals, and including larger sample sizes were generally related to an improved predictive performance. This review also highlights that very few studies (7/84, 8%) reported on challenges related to generalizability and less than half (38/84, 45%) of the studies reported on challenges related to explainability. Addressing these shortcomings will be instrumental in unlocking the full potential of DL for enhancing health care outcomes and patient care. PROSPERO CRD42018112161; https://tinyurl.com/yc6h9rwu.

The human microbiome is affected by multiple factors, including the environment and host genetics. In this study, we analyzed the salivary microbiomes of an extended family of Ashkenazi Jewish individuals living in several cities and investigated associations with both shared household and host genetic similarities. We found that environmental effects dominated over genetic effects. While there was weak evidence of geographical structuring at the level of cities, we observed a large and significant effect of shared household on microbiome composition, supporting the role of the immediate shared environment in dictating the presence or absence of taxa. This effect was also seen when including adults who had grown up in the same household but moved out prior to the time of sampling, suggesting that the establishment of the salivary microbiome earlier in life may affect its long-term composition. We found weak associations between host genetic relatedness and microbiome dissimilarity when using family pedigrees as proxies for genetic similarity. However, this association disappeared when using more-accurate measures of kinship based on genome-wide genetic markers, indicating that the environment rather than host genetics is the dominant factor affecting the composition of the salivary microbiome in closely related individuals. Our results support the concept that there is a consistent core microbiome conserved across global scales but that small-scale effects due to a shared living environment significantly affect microbial community composition. IMPORTANCE Previous research shows that the salivary microbiomes of relatives are more similar than those of nonrelatives, but it remains difficult to distinguish the effects of relatedness and shared household environment. Furthermore, pedigree measures may not accurately measure host genetic similarity. In this study, we include genetic relatedness based on genome-wide single nucleotide polymorphisms (SNPs) (rather than pedigree measures) and shared environment in the same analysis. We quantify the relative importance of these factors by studying the salivary microbiomes in members of a large extended Ashkenazi Jewish family living in different locations. We find that host genetics plays no significant role and that the dominant factor is the shared environment at the household level. We also find that this effect appears to persist in individuals who have moved out of the parental household, suggesting that aspects of salivary microbiome composition established during upbringing can persist over a time scale of years. Previous research shows that the salivary microbiomes of relatives are more similar than those of nonrelatives, but it remains difficult to distinguish the effects of relatedness and shared household environment. Furthermore, pedigree measures may not accurately measure host genetic similarity. In this study, we include genetic relatedness based on genome-wide single nucleotide polymorphisms (SNPs) (rather than pedigree measures) and shared environment in the same analysis. We quantify the relative importance of these factors by studying the salivary microbiomes in members of a large extended Ashkenazi Jewish family living in different locations. We find that host genetics plays no significant role and that the dominant factor is the shared environment at the household level. We also find that this effect appears to persist in individuals who have moved out of the parental household, suggesting that aspects of salivary microbiome composition established during upbringing can persist over a time scale of years.

PurposeTo develop and validate deep learning models predicting keratoconus progression risk using multimodal imaging, enabling risk-stratified monitoring.DesignRetrospective cohort study with internal and external validation.Participants7,396 eyes (3,893 patients) from MS-39 multimodal imaging, 963 eyes (519 patients) for external MS-39 validation, and 4,498 eyes (2,983 patients) with Pentacam tomography data.MethodsProgression was defined using global consensus criteria requiring changes in multiple parameters above device-specific precision limits. We compared conventional machine learning (XGBoost), unimodal deep learning, and multimodal fusion architectures for predicting two-year progression from baseline data. Recurrent neural networks (LSTM) incorporated sequential visit data. Clinical utility was assessed through simulated risk-stratified triage using predictive values.Main Outcome MeasuresAUROC, sensitivity, specificity, PPV, and NPV for progression prediction.ResultsThe MS-39 multimodal model (AS-OCT, Placido, tabular data) achieved AUROC 0.84 (95% CI: 0.83–0.85) at baseline, improving to 0.93 (95% CI: 0.91–0.96) with sequential LSTM models. Pentacam tabular data achieved AUROC 0.81 (95% CI: 0.78–0.84), improving to 0.87 (95% CI: 0.85–0.89) with two-visit LSTM. In simulated triage, 83% of patients were classified as low risk (NPV >90%) and 10% as high risk (PPV >90%) for consideration of early cross-linking.ConclusionsAI-driven risk stratification enables efficient monitoring across imaging platforms, with high NPV (>90%) safely reducing follow-up for low-risk patients and high PPV (>90%) ensuring timely intervention for high-risk cases.

To evaluate the feasibility of integrating genetic, imaging, and demographic data for predictive modelling of treatment outcomes in neovascular age-related macular degeneration (nAMD). Proof-of-concept retrospective cohort study with prospective DNA collection. Patients with unilateral nAMD receiving anti-vascular endothelial growth factor (anti-VEGF) therapy on a treat-and-extend regimen at a single tertiary centre were recruited. Polygenic risk scores (PRS) for AMD were derived from genotyping data (NIHR Bioresource). Optical coherence tomography (OCT) biomarkers-intraretinal fluid (IRF), subretinal fluid (SRF), pigment epithelial detachment (PED), and subretinal hyperreflective material (SHRM)-were automatically quantified using a deep learning segmentation model. Predictors of treatment outcomes included PRS, age at first injection, and OCT feature volumes at baseline. XGBoost was used for binary outcomes and linear regression for continuous outcomes, employing five-fold cross-validation. (1) macular dryness (no IRF/SRF) at 24 months, (2) average treatment interval in year 2, and (3) age at first injection. 106 participants were included. The multimodal model integrating age, imaging, and PRS predicted macular dryness at 24 months with AUC = 0.903, outperforming imaging alone (AUC = 0.701). PRS was associated with younger age at first injection (β = –4.69, 95% CI [–8.93, –0.44], P = 0.031) but not with treatment burden (β = –6.39, P = 0.13). Integrating PRS with OCT-derived imaging biomarkers and patient age is technically feasible and improves predictive performance of modelling for anatomical treatment outcomes in nAMD. PRS reflects genetic susceptibility to nAMD and contextualizes the predictive value of imaging biomarkers for treatment response.

BackgroundInherited retinal diseases (IRDs) are the leading cause of blindness in young people in the UK. Despite significant improvements in genomics medicine, the diagnosis of these conditions remains challenging, and around 40% do not receive a definite genetic diagnosis after extensive genetic testing. This survey aims to investigate the experience of individuals affected by IRDs, their relatives, friends and caregivers, focusing on their care and diagnostic journey. Additionally, it explores the potential acceptability of artificial intelligence (AI) technologies, such as Eye2Gene, that predict causative genes from retinal images of patients with IRDs.MethodsThis cross-sectional survey included Likert scale and open-ended questions and was distributed electronically using the Qualtrics platform between April and August 2024. The survey included questions on respondent demographics; their journey to receive specialist care and genetic testing; their information needs and their attitude towards AI-augmented diagnosis. Descriptive statistics and content analysis were used to interpret the survey responses.ResultsThe survey had 247 responses, of which 242 were analysed after removing four duplicates and one without consent; 80.2% were patients and the remainder were relatives, friends or caregivers. There was substantial variability in patient diagnostic journeys in terms of waiting times to see a specialist (IQR, 1–4 years), commute required (IQR, 10–74 miles) and number of visits to reach a diagnosis (IQR, 2–4). A substantial proportion of patients (35.8%) had a change in diagnosis. The majority of respondents (>90%) were overwhelmingly in favour of the integration of AI into the IRD pathway to accelerate genetic diagnosis and improve care.ConclusionThis survey identifies several key gaps and disparities in the IRD care pathway which may potentially be bridged with AI. The survey also reveals a favourable attitude towards incorporating AI into diagnostic testing of IRDs.