Artificial Intelligence Software for Diabetic Eye Screening: Diagnostic Performance and Impact of Stratification.

AIM: To evaluate the MONA.health artificial intelligence screening software for detecting referable diabetic retinopathy (DR) and diabetic macular edema (DME), including subgroup analysis. METHODS: The algorithm's threshold value was fixed at the 90% sensitivity operating point on the receiver operating curve to perform the disease classification. Diagnostic performance was appraised on a private test set and publicly available datasets. Stratification analysis was executed on the private test set considering age, ethnicity, sex, insulin dependency, year of examination, camera type, image quality, and dilatation status. RESULTS: The software displayed an area under the curve (AUC) of 97.28% for DR and 98.08% for DME on the private test set. The specificity and sensitivity for combined DR and DME predictions were 94.24 and 90.91%, respectively. The AUC ranged from 96.91 to 97.99% on the publicly available datasets for DR. AUC values were above 95% in all subgroups, with lower predictive values found for individuals above the age of 65 (82.51% sensitivity) and Caucasians (84.03% sensitivity). CONCLUSION: We report good overall performance of the MONA.health screening software for DR and DME. The software performance remains stable with no significant deterioration of the deep learning models in any studied strata.

Retinal vessel multifractals predict pial collateral status in patients with acute ischemic stroke.

OBJECTIVES: Pial collateral blood flow is a major determinant of the outcomes of acute ischemic stroke. This study was undertaken to determine whether retinal vessel metrics can predict the pial collateral status and stroke outcomes in patients. METHODS: Thirty-five patients with acute stroke secondary to middle cerebral artery (MCA) occlusion underwent grading of their pial collateral status from computed tomography angiography and retinal vessel analysis from retinal fundus images. RESULTS: The NIHSS (14.7 ± 5.5 vs 10.1 ± 5.8, p = 0.026) and mRS (2.9 ± 1.6 vs 1.9 ± 1.3, p = 0.048) scores were higher at admission in patients with poor compared to good pial collaterals. Retinal vessel multifractals: D0 (1.673±0.028vs1.652±0.025, p = 0.028), D1 (1.609±0.027vs1.590±0.025, p = 0.044) and f(α)max (1.674±0.027vs1.652±0.024, p = 0.019) were higher in patients with poor compared to good pial collaterals. Furthermore, support vector machine learning achieved a fair sensitivity (0.743) and specificity (0.707) for differentiating patients with poor from good pial collaterals. Age (p = 0.702), BMI (p = 0.422), total cholesterol (p = 0.842), triglycerides (p = 0.673), LDL (p = 0.952), HDL (p = 0.366), systolic blood pressure (p = 0.727), HbA1c (p = 0.261) and standard retinal metrics including CRAE (p = 0.084), CRVE (p = 0.946), AVR (p = 0.148), tortuosity index (p = 0.790), monofractal Df (p = 0.576), lacunarity (p = 0.531), curve asymmetry (p = 0.679) and singularity length (p = 0.937) did not differ between patients with poor compared to good pial collaterals. CONCLUSIONS: This is the first translational study to show increased retinal vessel multifractal dimensions in patients with acute ischemic stroke and poor pial collaterals. A retinal vessel classifier was developed to differentiate between patients with poor and good pial collaterals and may allow rapid non-invasive identification of patients with poor pial collaterals.

Age-related changes in the fractal dimension of the retinal microvasculature, effects of cardiovascular risk factors and smoking behaviour.

PURPOSE: This cross-sectional study investigates the association between retinal vessel complexity and age and studies the effects of cardiovascular health determinants. METHODS: Retinal vessel complexity was assessed by calculating the box-counting fractal dimension (Df ) from digital fundus photographs of 850 subjects (3-97 years). All photographs were labelled as 'non-pathological' by the treating ophthalmologist. RESULTS: Statistical models showed a significantly decreasing relationship between age and Df (linear: R-squared = 0.1897, p < 0.0001; quadratic: R-squared = 0.2343, p < 0.0001; cubic: R-squared = 0.2721, p < 0.0001), with the cubic regression model offering the best compromise between accuracy and model simplicity. Multivariate cubic regression showed that age, spherical equivalent and smoking behaviour have an effect (p < 0.0001) on Df . A significantly increasing effect of the number of pack-years on Df was observed (effect: 0.0004, p = 0.0017), as well as a significantly decreasing effect of years since tobacco abstinence (effect: -0.0149, p < 0.0001). CONCLUSION: We propose using a cubic trend with age, refractive error and smoking behaviour when interpreting retinal vessel complexity.

Synthetic Image Rendering Solves Annotation Problem in Deep Learning Nanoparticle Segmentation.

Nanoparticles occur in various environments as a consequence of man-made processes, which raises concerns about their impact on the environment and human health. To allow for proper risk assessment, a precise and statistically relevant analysis of particle characteristics (such as size, shape, and composition) is required that would greatly benefit from automated image analysis procedures. While deep learning shows impressive results in object detection tasks, its applicability is limited by the amount of representative, experimentally collected and manually annotated training data. Here, an elegant, flexible, and versatile method to bypass this costly and tedious data acquisition process is presented. It shows that using a rendering software allows to generate realistic, synthetic training data to train a state-of-the art deep neural network. Using this approach, a segmentation accuracy can be derived that is comparable to man-made annotations for toxicologically relevant metal-oxide nanoparticle ensembles which were chosen as examples. The presented study paves the way toward the use of deep learning for automated, high-throughput particle detection in a variety of imaging techniques such as in microscopies and spectroscopies, for a wide range of applications, including the detection of micro- and nanoplastic particles in water and tissue samples.

Retinal microvascular complexity comparing mono- and multifractal dimensions in relation to cardiometabolic risk factors in a Middle Eastern population.

PURPOSE: Metrics that capture changes in the retinal microvascular structure are relevant in the context of cardiometabolic disease development. The microvascular topology is typically quantified using monofractals, although it obeys more complex multifractal rules. We study mono- and multifractals of the retinal microvasculature in relation to cardiometabolic factors. METHODS: The cross-sectional retrospective study used data from 3000 Middle Eastern participants in the Qatar Biobank. A total of 2333 fundus images (78%) passed quality control and were used for further analysis. The monofractal (Df ) and five multifractal metrics were associated with cardiometabolic factors using multiple linear regression and were studied in clinically relevant subgroups. RESULTS: Df and multifractals are lowered in function of age, and Df is lower in males compared to females. In models corrected for age and sex, Df is significantly associated with BMI, insulin, systolic blood pressure, glycated haemoglobin (HbA1c), albumin, LDL and total cholesterol concentrations. Multifractals are negatively associated with systolic and diastolic blood pressure, glucose and the WHO/ISH cardiovascular risk score. Df was higher, and multifractal curve asymmetry was lower in diabetic patients (HbA1c > 6.5%) compared to healthy individuals (HbA1c < 5.7%). Insulin resistance (insulin ≥ 23 mcU/mL) was associated with significantly lower Df values. CONCLUSION: One or more fractal metrics are in association with sex, age, BMI, systolic and diastolic blood pressure and biochemical blood measurements in a Middle Eastern population study. Follow-up studies aiming at investigating retinal microvascular changes in relation to cardiometabolic risk should analyse both monofractal and multifractal metrics for a more comprehensive microvascular picture.

Association of Retinal Microvascular Characteristics With Short-term Memory Performance in Children Aged 4 to 5 Years.

Importance: Neurocognitive functions develop rapidly in early childhood and depend on the intrinsic cooperation between cerebral structures and the circulatory system. The retinal microvasculature can be regarded as a mirror image of the cerebrovascular circulation. Objective: To investigate the association between retinal vessel characteristics and neurological functioning in children aged 4 to 5 years. Design, Setting, and Participants: In this cohort study, mother-child pairs were recruited at birth from February 10, 2010, to June 24, 2014, and renewed consent at their follow-up visit from December 10, 2014, to July 13, 2018. Participants were followed up longitudinally within the prospective Environmental Influence on Aging in Early Life birth cohort. A total of 251 children underwent assessment for this study. Data were analyzed from July 17 to October 30, 2019. Main Outcomes and Measures: Retinal vascular diameters, the central retinal arteriolar equivalent (CRAE), central retinal venular equivalent (CRVE), vessel tortuosity, and fractal dimensions were determined. Attention and psychomotor speed, visuospatial working memory, and short-term visual recognition memory were assessed by the Cambridge Neuropsychological Test Automated Battery, including the following tasks: Motor Screening (MOT), Big/Little Circle (BLC), Spatial Span (SSP), and Delayed Matching to Sample (DMS). Results: Among the 251 children included in the assessment (135 girls [53.8%]; mean [SD] age, 4.5 [0.4] years), for every 1-SD widening in CRVE, the children performed relatively 2.74% (95% CI, -0.12 to 5.49; P = .06) slower on the MOT test, had 1.76% (95% CI, -3.53% to -0.04%; P = .04) fewer correct DMS assessments in total, and made 2.94% (95% CI, 0.39 to 5.29; P = .02) more errors given a previous correct answer in the DMS task on multiple linear regression modeling. For every 1-SD widening in CRAE, the total percentage of errors and errors given previous correct answers in the DMS task increased 1.44% (95% CI, -3.25% to 0.29%; P = .09) and 2.30% (95% CI, -0.14% to 4.61%; P = .07), respectively. A 1-SD higher vessel tortuosity showed a 4.32% relative increase in latency in DMS task performance (95% CI, -0.48% to 9.12%; P = .07). Retinal vessel characteristics were not associated with BLC and SSP test outcomes. Conclusions and Relevance: These findings suggest that children's microvascular phenotypes are associated with short-term memory and that changes in the retinal microvasculature may reflect neurological development during early childhood.

Age and sex affect deep learning prediction of cardiometabolic risk factors from retinal images.

Deep neural networks can extract clinical information, such as diabetic retinopathy status and individual characteristics (e.g. age and sex), from retinal images. Here, we report the first study to train deep learning models with retinal images from 3,000 Qatari citizens participating in the Qatar Biobank study. We investigated whether fundus images can predict cardiometabolic risk factors, such as age, sex, blood pressure, smoking status, glycaemic status, total lipid panel, sex steroid hormones and bioimpedance measurements. Additionally, the role of age and sex as mediating factors when predicting cardiometabolic risk factors from fundus images was studied. Predictions at person-level were made by combining information of an optic disc centred and a macula centred image of both eyes with deep learning models using the MobileNet-V2 architecture. An accurate prediction was obtained for age (mean absolute error (MAE): 2.78 years) and sex (area under the curve: 0.97), while an acceptable performance was achieved for systolic blood pressure (MAE: 8.96 mmHg), diastolic blood pressure (MAE: 6.84 mmHg), Haemoglobin A1c (MAE: 0.61%), relative fat mass (MAE: 5.68 units) and testosterone (MAE: 3.76 nmol/L). We discovered that age and sex were mediating factors when predicting cardiometabolic risk factors from fundus images. We have found that deep learning models indirectly predict sex when trained for testosterone. For blood pressure, Haemoglobin A1c and relative fat mass an influence of age and sex was observed. However, achieved performance cannot be fully explained by the influence of age and sex. In conclusion we confirm that age and sex can be predicted reliably from a fundus image and that unique information is stored in the retina that relates to blood pressure, Haemoglobin A1c and relative fat mass. Future research should focus on stratification when predicting person characteristics from a fundus image.

Children's microvascular traits and ambient air pollution exposure during pregnancy and early childhood: prospective evidence to elucidate the developmental origin of particle-induced disease.

BACKGROUND: Particulate matter exposure during in utero life may entail adverse health outcomes later in life. The microvasculature undergoes extensive, organ-specific prenatal maturation. A growing body of evidence shows that cardiovascular disease in adulthood is rooted in a dysfunctional fetal and perinatal development, in particular that of the microcirculation. We investigate whether prenatal or postnatal exposure to PM2.5 (particulate matter with a diameter ≤ 2.5 µm) or NO2 is related to microvascular traits in children between the age of four and six. METHODS: We measured the retinal microvascular diameters, the central retinal arteriolar equivalent (CRAE) and central retinal venular equivalent (CRVE), and the vessel curvature by means of the tortuosity index (TI) in young children (mean [SD] age 4.6 [0.4] years), followed longitudinally within the ENVIRONAGE birth cohort. We modeled daily prenatal and postnatal PM2.5 and NO2 exposure levels for each participant's home address using a high-resolution spatiotemporal model. RESULTS: An interquartile range (IQR) increase in PM2.5 exposure during the entire pregnancy was associated with a 3.85-µm (95% CI, 0.10 to 7.60; p = 0.04) widening of the CRVE and a 2.87-µm (95% CI, 0.12 to 5.62; p = 0.04) widening of the CRAE. For prenatal NO2 exposure, an IQR increase was found to widen the CRVE with 4.03 µm (95% CI, 0.44 to 7.63; p = 0.03) and the CRAE with 2.92 µm (95% CI, 0.29 to 5.56; p = 0.03). Furthermore, a higher TI score was associated with higher prenatal NO2 exposure. We observed a postnatal effect of short-term PM2.5 exposure on the CRAE and a childhood NO2 exposure effect on both the CRVE and CRAE. CONCLUSIONS: Our results link prenatal and postnatal air pollution exposure with changes in a child's microvascular traits as a fundamental novel mechanism to explain the developmental origin of cardiovascular disease.

Systematic Comparison of Heatmapping Techniques in Deep Learning in the Context of Diabetic Retinopathy Lesion Detection.

Purpose: Heatmapping techniques can support explainability of deep learning (DL) predictions in medical image analysis. However, individual techniques have been mainly applied in a descriptive way without an objective and systematic evaluation. We investigated comparative performances using diabetic retinopathy lesion detection as a benchmark task. Methods: The Indian Diabetic Retinopathy Image Dataset (IDRiD) publicly available database contains fundus images of diabetes patients with pixel level annotations of diabetic retinopathy (DR) lesions, the ground truth for this study. Three in advance trained DL models (ResNet50, VGG16 or InceptionV3) were used for DR detection in these images. Next, explainability was visualized with each of the 10 most used heatmapping techniques. The quantitative correspondence between the output of a heatmap and the ground truth was evaluated with the Explainability Consistency Score (ECS), a metric between 0 and 1, developed for this comparative task. Results: In case of the overall DR lesions detection, the ECS ranged from 0.21 to 0.51 for all model/heatmapping combinations. The highest score was for VGG16+Grad-CAM (ECS = 0.51; 95% confidence interval [CI]: [0.46; 0.55]). For individual lesions, VGG16+Grad-CAM performed best on hemorrhages and hard exudates. ResNet50+SmoothGrad performed best for soft exudates and ResNet50+Guided Backpropagation performed best for microaneurysms. Conclusions: Our empirical evaluation on the IDRiD database demonstrated that the combination DL model/heatmapping affects explainability when considering common DR lesions. Our approach found considerable disagreement between regions highlighted by heatmaps and expert annotations. Translational Relevance: We warrant a more systematic investigation and analysis of heatmaps for reliable explanation of image-based predictions of deep learning models.