Retinal Images Research Articles

Multi-task OCTA image segmentation with innovative dimension compression

Optical Coherence Tomography Angiography (OCTA) plays a crucial role in the early detection and continuous monitoring of ocular diseases, which relies on accurate multi-tissue segmentation of retinal images. Existing OCTA segmentation methods typically focus on single-task designs that do not fully utilize the information of volume data in these images. To bridge this gap, our study introduces H2C-Net, a novel network architecture engineered for simultaneous and precise segmentation of various retinal structures, including capillaries, arteries, veins, and the fovea avascular zone (FAZ). At its core, H2C-Net consists of a plug-and-play Height-Channel Module (H2C) and an Enhanced U-shaped Network (GPC-Net). The H2C module cleverly converts the height information of the OCTA volume data into channel information through the Squeeze operation, realizes the lossless dimensionality reduction from 3D to 2D, and provides the "Soft layering" information by unidirectional pooling. Meanwhile, in order to guide the network to focus on channels for training, U-Net is enhanced with group normalization, channel attention mechanism, and Parametric Rectified Linear Unit (PReLU), which reduces the dependence on batch size and enhances the network's ability to extract salient features. Extensive experiments on two subsets of the publicly available OCTA-500 dataset have shown that H2C-Net outperforms existing state-of-the-art methods. It achieves average Intersection over Union (IoU) scores of 82.84 % and 88.48 %, marking improvements of 0.81 % and 1.59 %, respectively. Similarly, the average Dice scores are elevated to 90.40 % and 93.76 %, exceeding previous benchmarks by 0.42 % and 0.94 %. The proposed H2C-Net exhibits excellent performance in OCTA image segmentation, providing an efficient and accurate multi-task segmentation solution in ophthalmic diagnostics. The code is publicly available at: https://github.com/IAAI-SIT/H2C-Net.

HyMNet: A Multimodal Deep Learning System for Hypertension Prediction Using Fundus Images and Cardiometabolic Risk Factors

Study Objectives: This study aimed to develop a multimodal deep learning (MMDL) system called HyMNet, integrating fundus images and cardiometabolic factors (age and sex) to enhance hypertension (HTN) detection. Methods: HyMNet employed RETFound, a model pretrained on 1.6 million retinal images, for the fundus data, in conjunction with a fully connected neural network for age and sex. The two pathways were jointly trained by joining their feature vectors into a fusion network. The system was trained on 5016 retinal images from 1243 individuals provided by the Saudi Ministry of National Guard Health Affairs. The influence of diabetes on HTN detection was also assessed. Results: HyMNet surpassed the unimodal system, achieving an F1 score of 0.771 compared to 0.745 for the unimodal model. For diabetic patients, the F1 score was 0.796, while it was 0.466 for non-diabetic patients. Conclusions: HyMNet exhibited superior performance relative to unimodal approaches, with an F1 score of 0.771 for HyMNet compared to 0.752 for models trained on demographic data alone, underscoring the advantages of MMDL systems in HTN detection. The findings indicate that diabetes significantly impacts HTN prediction, enhancing detection accuracy among diabetic patients. Utilizing MMDL with diverse data sources could improve clinical applicability and generalization.

Detection of optic disc in human retinal images utilizing the Bitterling Fish Optimization (BFO) algorithm

Early detection and correct identification of the optic disc (OD) on scanned retinal images are significant for diagnosing and treating several ophthalmic conditions, including glaucoma and diabetic retinopathy. Conventional methods for detecting the OD often struggle with processing retinal images due to noise, changes in illumination, and complex overlapping images. This study presents the development of effective and accurate fixation of the optic disc using the Bitterling Fish Optimization (BFO) algorithm, which enhances the processes of OD imaging in speed and precision. The proposed method begins with image enhancement and noise suppression for preprocessing, followed by applying the BFO algorithm to locate and delineate the OD region. The performance evaluation of the algorithm was conducted within several public domain retinal images, including DRIVE, STARE, ORIGA, DRISHTI-GS, DiaRetDB0, and DiaRetDB1 datasets about some internal metrics: sensitivity (SE), specificity (SP), accuracy (ACC), DICE overlap coefficient, overlap and time of processing respectively. The technique based on BFO provided better results, with 99.33%, 99.94%, and 98.22% accuracy achieved for OD in DRIVE, DRISHTI-GS, and DiaRetDB 1, respectively. The approach also demonstrated high overlaps and good DICE results, with a DICE coefficient of 0.9501 for the DRISHTI-GS database. On average, the processing time per image was less than 2.5 s, proving the approach’s efficiency in computations. The BFO approach has demonstrated its effectiveness and scalability in detecting optic discs in retinal images in an automated manner. It showed impressive performance levels in terms of computation time and accuracy and was variation resistant irrespective of the quality of the image and the pathology present on it. This method holds significant potential for clinical use, providing a meaningful way of diagnosing and managing ocular disease at an early stage.

Deep learning detection of subclinical cardiovascular dysfunction in type 2 diabetes through retinal photography

Abstract Background Type 2 diabetes (T2D) is associated with excess atherosclerotic coronary disease and heart failure risk, with several cardiac abnormalities described in asymptomatic patients. Routine digital retinal photography performed for screening of diabetic retinopathy (DR) has the potential to identify patients with underlying cardiovascular disease (CVD). We aimed to determine whether features extracted from digital retinal photographs could detect subclinical cardiovascular dysfunction in asymptomatic people with T2D. Methods We prospectively enrolled adults with T2D and no history or symptoms of CVD to undergo extensive phenotyping with multiparametric stress perfusion cardiac MRI and CT coronary artery calcium scoring. Digital retinal photographs acquired for routine diabetic eye screening and performed within one year of cardiac evaluation were interrogated for retinopathy grading and using four deep learning (DL) models (Xception, Inception, EfficientNet, and MobileNet) to capture complex patterns within the retinal vasculature. Each subject had four retinal photographs (two of each eye) for DL analysis. Cardiac image analysis was performed blinded to participant details and independent of retinal images. Subjects with and without DR were compared. Participants were divided into training (80%) and validation datasets (20%) prior to DL analysis. DL models employed to predict from retinal images: presence or absence of coronary artery calcium on non-contrast CT, myocardial perfusion reserve <2.5, left ventricular (LV) global longitudinal strain <16%, absolute stress and rest myocardial blood flood (mL/g/min) used as continuous variables. AUC was used to assess the ability of DL models to discriminate between those without and without subclinical CVD. Results An initial 254 subjects (comprising 1,016 retinal photographs) were included in the analysis: 212 (83%) with no DR, 42 (17%) with grade 1 (mild background) DR. Key cardiac imaging parameters in subjects with and without DR are displayed in Table 1. The groups were well matched for age, sex, body mass index and blood pressure. Overall, those with grade 1 DR had higher atheroma burden, evidence of increased LV filling pressures (E/e’) and poorer systolic function, but no differences in stress and rest myocardial blood flow or myocardial perfusion reserve. However, all four DL models failed to achieve sufficient discrimination of any of the evaluated cardiac imaging parameters, with the majority achieving AUC scores below 50% for all cardiac parameters. Conclusion Asymptomatic adults with grade 1 DR exhibit more subclinical atherosclerosis and cardiovascular dysfunction than those without DR. However, DL transfer models could not discriminate between subjects with and without cardiovascular disease. Further work is needed to better define whether retinal screening may be used to identify those people with T2D most at risk of underlying cardiovascular disease.Subjects demographics

Self-supervised based clustering for retinal optical coherence tomography images.

In response to the inadequacy of manual analysis in meeting the rising demand for retinal optical coherence tomography (OCT) images, a self-supervised learning-based clustering model was implemented. A public dataset was utilized, with 83,484 OCT images with categories of choroidal neovascularization (CNV), diabetic macular edema (DME), drusen, and normal fundus. This study employed the Semantic Pseudo Labeling for Image Clustering (SPICE) framework, a self-supervised learning-based method, to cluster unlabeled OCT images into binary and four categories, and the performances were compared with baseline models. We also analysed feature distribution using t-SNE, and explored the cluster centers, attention maps, and misclassified images. In addition, DME and CNV subsets were clustered binarily, and the results were interpreted by two retinal specialists. SPICE demonstrated superior performance in binary and four categories classification tasks, achieving the accuracy of 0.886 and 0.846, respectively. In t-SNE analysis, the four types exhibited significant clustering into distinct groups. The cluster centers corresponded to the human labels, and the heat map revealed that the model focused on important biomarkers. The misclassified images exposed similar features to the inaccurate classes. The model also grouped DME and CNV into two distinct categories respectively. Self-supervised clustering effectively distinguished disease variances and revealed common features, with a notable capability to detect disease heterogeneity through biomarkers.

Revolutionising Diabetic Retinopathy Diagnosis with Modified Regularisation Long Short-Term Memory Framework

The diagnosis of Diabetic Retinopathy (DR) demands a paradigm shift towards more accurate and efficient solutions to overcome vision impairment. Therefore, the current study introduces a new Modified Regularisation Long Short-term Memory (MR-LSTM) framework approach for DR diagnosis. The proposed framework leverages the power of deep learning and provides a dynamic and robust solution for the early detection of DR, which in turn preserves a patient’s vision. The proposed framework uses a DR Debrecen Dataset from the UCI database with 21 distinct features relevant to retinal health, and employs a series of data preprocessing steps, including data cleaning, normalisation, and transformation, to ensure data quality and compatibility. The MR-LSTM framework excels at capturing temporal dependencies in sequential retinal images, offering a unique advantage in understanding the progression of DR. The MR-LSTM framework is implemented using Python libraries, and the results are compared with those of other popular models. It is observed that the MR-LSTM framework outperforms other models and achieves an accuracy of 97.12 percent and an F1 Score of 98.49. Furthermore, the Receiver Operating Characteristic (ROC) curve reveals an area under the curve of 0.97, highlighting the exceptional ability to discriminate between positive and negative cases of the proposed framework. By revolutionising DR diagnosis with the proposed MR-LSTM framework, it can achieve accurate, timely, and accessible solutions in the fight against vision-threatening conditions.

Widefield optical coherence tomography by electro-optical modulation

Optical coherence tomography (OCT) is a unique imaging modality capable of axial sectioning with a resolution of only a few microns. Its ability to image with high resolution deep within tissue makes it ideal for material inspection, dentistry, and, in particular, ophthalmology. Widefield retinal imaging has garnered increasing clinical interest for the detection of numerous retinal diseases. However, real-time applications in clinical practice demand the contrast of swept-source OCT at scan speeds that limit their depth range. The curvature of typical samples, such as teeth, corneas, or retinas, thus restricts the field-of-view of fast OCT systems. Novel high-speed swept sources are expected to further improve the scan rate; however, not without exacerbating the already severe trade-off in depth range. Here, we show how, without the need for mechanical repositioning, harmonic images can be rapidly synthesized at any depth. This is achieved by opto-electronic modulation of a single-frequency swept source laser in tandem with tailored numerical dispersion compensation. We demonstrate experimentally how real-time imaging of highly-curved samples is enabled by extending the effective depth-range 8-fold. Even at the scan speed of a 400 kHz swept source, harmonic OCT enables widefield retinal imaging.

Astigmatism and its Associated Factors in Pediatric Population, Nepal

Introduction: Astigmatism is a common vision disorder that leads to blurred vision due to the inability of the ocular system to form a sharply focused image on the retina. A child with uncorrected astigmatism experiences blur on a continuous basis and also child has a clear retinal image when looking at a near environment. Objective: The objective of this study was to access the prevalence and risk factors associated with Astigmatism in pediatric population. Method: This was a hospital based cross sectional study conducted among 2 to 14 years children with astigmatism of ≥ 1.0 D in either eye. Study conducted from 1st September 2022 to 28th February 2023. Data was entered in Ms-Excel and further analyzed in SPSS Version 12.0. Result: Prevalence of astigmatism was found to be 38.80%. Prevalence in Male was found to be higher 65.20% than in female. Age was found to be statistically significant p = (0.003) with astigmatism. Conclusion: Astigmatism is a common vision disorder that leads to blurred vision due to the inability of the ocular system to form a sharply focused image on the retina. Concerned stakeholder need to screen the problem to minimize preventable blindness in children.

Diabetic Retinopathy Detection using CNN and Telemedicine

Diabetic Retinopathy (DR) is a leading cause of vision impairment and blindness among individuals with diabetes. Early detection and timely intervention are critical for preventing severe complications. However, the scarcity of ophthalmologists, especially in rural and underdeveloped regions, limits the accessibility of timely diagnosis and treatment. Telemedicine, combined with advanced artificial intelligence (AI) techniques, such as Convolutional Neural Networks (CNN), offers a promising solution for improving DR detection and patient care remotely. This paper presents a comprehensive survey of recent advancements in DR detection using CNN-based deep learning models, along with telemedicine frameworks for real-time consultation and treatment suggestions. We discuss the underlying architecture of CNN models and their role in automating retinal image classification.

The Association between Intermittent Fasting and Glaucoma.

To examine the association between intermittent fasting ("breakfast skipping") and glaucoma in the National Health and Nutrition Examination Survey (NHANES). Retrospective, cross-sectional study of adult participants of the 2005-2008 NHANES aged 40 years and older. Intermittent fasting was defined as skipping breakfast on both days of the NHANES dietary interview. Glaucoma was defined by 1) self-reported glaucoma diagnosis, 2) cup-to-disc (CDR) ≥0.6 in at least one eye, and 3) Possible, probable or definite glaucoma based on clinical judgment of retinal imaging of individuals with CDR ≥0.6 in at least one eye. Covariates included age, sex, race/ethnicity, body mass index, blood pressure, and waist circumference. Chi-Square and Fisher Exact Test were used for categorical variables and t-test for continuous variables. All data were weighted based on the NHANES multistage sampling design. The population included 7,081 individuals ≥40 years old, of whom 946 (13.4%) reported skipping breakfast on both interview days; 482 (6.8%) individuals self-reported glaucoma diagnosis, 549 (7.8%) individuals had glaucoma based on CDR ≥0.6 and 343 (6.0%) individuals had glaucoma based on clinical judgment. Diagnosis of glaucoma by all definitions was associated with older age and Black or Latinx/Hispanic ethnicity (P=0.004) but was not associated with intermittent fasting status (P=0.151). In the 2005-2008 NHANES population, intermittent fasting, defined as breakfast skipping, was not associated with decreased risk of glaucoma diagnosis overall. Further studies are needed to examine the potential effects of intermittent fasting on glaucoma risk.

Accommodating version of a schematic eye for emmetropia and myopia.

To develop an accommodating, wide-angle, schematic eye for emmetropia and myopia in which spectacle refraction and accommodation level are input parameters. The schematic eye is based on an earlier unaccommodated refraction-dependent eye for myopia developed by Atchison in 2006. This has a parabolic gradient index lens and parameters derived from biometric and optical measurements on young adults. Several parameters are linearly dependent upon spectacle refraction (anterior radius of curvature of the cornea, axial length and vertex radii of curvature and conic asphericities of a biconic retina). The new accommodated schematic eye incorporates accommodation-dependent changes in several lens-related parameters. These changes are based on literature values for anterior chamber depth, lens thickness, vitreous chamber depth, lens surface radii of curvature and lens front surface asphericity. A parabolic variation of refractive index with relative distance from the lens centre is retained, with the same edge and centre refractive indices as the earlier model, but the distribution has been manipulated to maintain focus near the retina for the emmetropic case at 0 and 4 D accommodation. The asphericity of the lens back surface is changed so that spherical aberration and peripheral refraction approximately match typical literature trends. The model is used to compare spherical aberration and peripheral refraction in eyes with up to 4 D of myopia and 4 D of accommodation. The levels of spherical aberration in the unaccommodated schematic eyes are similar to literature values for young adults, but the changes in spherical aberration with accommodation are approximately two-thirds of that found in an experimental study. As intended, peripheral refractions in the accommodated schematic eyes are similar to those of their unaccommodated counterparts. The wide-angle model extends the range of schematic eyes to include both refraction and accommodation as variable input parameters. It may be useful in predicting aspects of retinal image quality.

Systematic Review and Meta-Analysis on the Association Between Daily Niacin Intake and Glaucoma

Background: Glaucoma is a progressive optic neuropathy, characterised by a complex pathophysiology, with mitochondrial dysfunction playing a significant role in the cellular damage and apoptosis of ganglion cells. Niacin is a precursor to several molecules acting as coenzymes in the mitochondrial production of ATP, in DNA repair and in the reduction of reactive oxygen species. The objective of this systematic review is to assess the impact of daily niacin intake on glaucoma. Methods: Case–control and cohort studies regarding niacin and glaucoma, indexed in PubMed, Web of Science, Cochrane and Scopus, were included. Other study methodologies, studies regarding niacin in other ocular disease or other nutrients in glaucoma were excluded. Bias was assessed using the Newcastle–Ottawa Scale. The study protocol was registered in the PROSPERO database (no. CRD42024578889). Results: Five case–control studies were included. In the pooled analysis, a significantly higher proportion of patients with high niacin consumption was found in the group without glaucoma compared to those with glaucoma as defined by ISGEO criteria (p-value < 0.00001; OR = 0.66, 95% CI 0.55–0.79) or as defined by retinal imaging (p-value = 0.02; OR = 0.63, 95% CI 0.43–0.94). Conclusions: Daily dietary intake of niacin is significantly lower in patients with glaucoma compared to the general population. Given different average daily intakes of niacin in these populations, different glaucoma definitions and several confounding variables which weaken the associations, large sample, standardised randomised controlled trials are needed to confirm the potential benefits of niacin in glaucoma.

ViT-HHO: Optimized vision transformer for diabetic retinopathy detection using Harris Hawk optimization

Diabetic retinopathy (DR) is a significant cause of vision impairment globally, emphasizing the importance of timely and precise detection to prevent severe consequences. This study presents an optimized Vision Transformer (ViT) model that incorporates Harris Hawk Optimization (HHO) to improve the automated detection of diabetic retinopathy (DR). The ViT architecture utilizes self-attention mechanisms to capture local and global features in retinal images. Additionally, HHO optimizes key hyperparameters to maximize the performance of the model. The proposed ViT-HHO model achieved exceptional performance on the APTOS-2019 and IDRiD datasets. Specifically, it achieved 99.83 % accuracy, 99.78 % sensitivity, 99.85 % specificity, and 99.80 % AUC-ROC on the APTOS-2019 dataset, surpassing traditional CNNs and alternative optimization techniques. The model exhibited strong generalization on the IDRiID dataset, achieving an accuracy of 99.11 % and an AUC-ROC of 99.12 %. The ViT-HHO model demonstrates the potential for enhancing the clinical detection of diabetic retinopathy (DR), providing high precision and reliability.•An optimized Vision Transformer (ViT) model was developed using HHO for improved detection of Diabetic Retinopathy (DR).•The model was validated on the APTOS-2019 and IDRiID datasets, demonstrating superior accuracy and AUC-ROC metrics.•The model's generalization and robustness were demonstrated through comprehensive performance evaluations.

Impact of Genetic and Environmental Factors on Peripheral Refraction.

Investigate genetic and environmental influences on refractive errors in monozygotic (MZ) and dizygotic (DZ) twin pairs. We assessed foveal and peripheral refractions in 54 MZ and 46 DZ twins, capturing three scans across the retina. The study focused on spherical equivalent (M) at the fovea (MLOS) and changes in midperipheral (δMmid-periphery), and peripheral (δMperiphery) defocus, along with nasal-temporal asymmetry (root mean squared error [RMSEASY]) and image shell contour (RMSEAVG). Genetic and environmental contributions were analyzed using structural equation models. No significant differences were observed between MZ and DZ twins for the examined variables. Intraclass correlations (ICC) indicated an important difference in genetic influence between MLOS, with the MZ twin pairs showing a higher correlation (0.83) than DZ (0.69) pairs, and δMperiphery, because the ICC for the MZ doubled (0.87) that of the DZ (0.42) pairs. Heritability estimates from the ACE model confirmed the large difference on genetic factors' influence on the variance for MLOS (0.13) and δMperiphery (0.77) change in refractive error. RMSEASY and RMSEAVG metrics showed significant genetic impact, particularly pronounced in the peripheral measurements, revealing high genetic control. The study delineates a marked environmental impact on central refractive errors, whereas genetic factors had a more significant influence on peripheral refractive variance and retinal image traits. Findings of the ACE model highlight the intricate genetic and environmental interplay in refractive error development, with a notable genetic dominance in peripheral vision characteristics. This suggests potential genetic targets for interventions in myopia management and emphasizes the need for personalized approaches based on genetic predispositions. Understanding the impact of genetics and environment on peripheral refraction is essential for deepening our fundamental knowledge of myopia and guiding the development of advanced myopia control strategies.

Artificial intelligence for detection of retinal toxicity in chloroquine and hydroxychloroquine therapy using multifocal electroretinogram waveforms.

Chloroquine and hydroxychloroquine, while effective in rheumatology, pose risks of retinal toxicity, necessitating regular screening to prevent visual disability. The gold standard for screening includes retinal imaging and automated perimetry, with multifocal electroretinography (mfERG) being a recognized but less accessible method. This study explores the efficacy of Artificial Intelligence (AI) algorithms for detecting retinal damage in patients undergoing (hydroxy-)chloroquine therapy. We analyze the mfERG data, comparing the performance of AI models that utilize raw mfERG time-series signals against models using conventional waveform parameters. Our classification models aimed to identify maculopathy, and regression models were developed to predict perimetric sensitivity. The findings reveal that while regression models were more adept at predicting non-disease-related variation, AI-based models, particularly those utilizing full mfERG traces, demonstrated superior predictive power for disease-related changes compared to linear models. This indicates a significant potential to improve diagnostic capabilities, although the unbalanced nature of the dataset may limit some applications.

Application of optical coherence tomography in neurodegenerative diseases: a focus on Parkinson’s, Alzheimer’s, and Schizophrenia in Bulgarian patients

This study investigates retinal alterations in patients with Alzheimer’s disease (AD), Parkinson’s disease (PD), and schizophrenia (SZ) using Optical Coherence Tomography (OCT). The primary objective was to determine whether these neurodegenerative diseases manifest in measurable retinal changes and to assess the impact of disease duration, medication use, and cognitive decline on these alterations. A cross-sectional observational design was employed, including 132 patients and age- and gender-matched controls. OCT imaging was performed using the Topcon 3D OCT-1 Maestro 2, focusing on the retinal nerve fibre layer (RNFL), ganglion cell complex (GCC), and macular thickness. Significant retinal thinning was observed in the AD and PD groups, correlating with disease severity and cognitive decline, and was more pronounced with longer disease duration. In contrast, no significant retinal changes were identified in the SZ group. The study also explored the effects of different drug classes, revealing correlations between specific medications and retinal parameters, particularly in the AD and PD cohorts. To our knowledge, this is the first study of its kind conducted with Bulgarian patients. These findings suggest that OCT may serve as a non-invasive biomarker for neurodegeneration in AD and PD, though its utility in SZ remains limited. Future research should aim to standardize OCT protocols, further investigate the potential of retinal imaging in tracking neurodegenerative disease progression, and explore its integration with other neuroimaging techniques for a more comprehensive diagnostic and monitoring approach.

Erratum — Optimized Improved Random Forest-Fostered Glaucoma Detection from Fundus Retinal Images

Erratum — Optimized Improved Random Forest-Fostered Glaucoma Detection from Fundus Retinal Images

OPTICAL BLUR AFFECTS DIFFERENTLY ON AND OFF VISUAL PATHWAYS.

Optical blur affects ON and OFF pathways differently.Blur expands ON but shrinks OFF receptive fields, weakening OFF more than ON signals.Magnification from positive blur reduces OFF more than ON signal strength and latency.ON-OFF response balance can signal retinal image quality and guide eye growth. Pons et al. demonstrate that optical blur affects differently the receptive field properties and responses from ON and OFF visual pathways, revealing a new relation between ON/OFF pathway balance and blurred visual perception. The results highlight a new role of ON and OFF pathways in optimizing retinal image quality that can have implications for the regulation of eye growth.

Early detection of dementia through retinal imaging and trustworthy AI

Alzheimer’s disease (AD) is a global healthcare challenge lacking a simple and affordable detection method. We propose a novel deep learning framework, Eye-AD, to detect Early-onset Alzheimer’s Disease (EOAD) and Mild Cognitive Impairment (MCI) using OCTA images of retinal microvasculature and choriocapillaris. Eye-AD employs a multilevel graph representation to analyze intra- and inter-instance relationships in retinal layers. Using 5751 OCTA images from 1671 participants in a multi-center study, our model demonstrated superior performance in EOAD (internal data: AUC = 0.9355, external data: AUC = 0.9007) and MCI detection (internal data: AUC = 0.8630, external data: AUC = 0.8037). Furthermore, we explored the associations between retinal structural biomarkers in OCTA images and EOAD/MCI, and the results align well with the conclusions drawn from our deep learning interpretability analysis. Our findings provide further evidence that retinal OCTA imaging, coupled with artificial intelligence, will serve as a rapid, noninvasive, and affordable dementia detection.

Heatmap analysis for artificial intelligence explainability in diabetic retinopathy detection: illuminating the rationale of deep learning decisions.

The opaqueness of artificial intelligence (AI) algorithms decision processes limit their application in healthcare. Our objective was to explore discrepancies in heatmaps originated from slightly different retinal images from the same eyes of individuals with diabetes, to gain insights into the deep learning (DL) decision process. Pairs of retinal images from the same eyes of individuals with diabetes, composed of images obtained before and after pupil dilation, underwent automatic analysis by a convolutional neural network for the presence of diabetic retinopathy (DR), output being a score ranging from 0 to 1. Gradient-based Class Activation Maps (GradCam) allowed visualization of activated areas. Pairs of images with discordant DL scores or outputs within the pair were objectively compared to the concordant pairs, regarding the sum of activations of Class Activation Mapping (CAM), the number of activated areas, and DL score differences. Heatmaps of discordant pairs were also qualitatively assessed. Algorithmic performance for the detection of DR attained 89.8% sensitivity, 96.3% specificity and area under the receiver operating characteristic (ROC) curve of 0.95. Out of 210 comparable pairs of images, 20 eyes and 10 eyes were considered discordant according to DL score difference and regarding DL output, respectively. Comparison of concordant versus discordant groups showed statistically significant differences for all objective variables. Qualitative analysis pointed to subtle differences in image quality within discordant pairs. The successfully established relationship among objective parameters extracted from heatmaps and DL output discrepancies reinforces the role of heatmaps for DL explainability, fostering acceptance of DL systems for clinical use.