Computer vision is a powerful tool in medical image analysis, supporting the early detection and classification of eye diseases. Diabetic retinopathy (DR), a severe eye disease secondary to diabetes, accompanies several early signs of eye-threatening conditions, such as microaneurysms (MAs), hemorrhages (HEMOs), and exudates (EXs), which have been widely studied and targeted as objects to be detected by computer vision models. In this work, we tested the performances of the state-of-the-art YOLOv8 and YOLOv9 architectures on DR fundus features segmentation without coding experience or a programming background. We took one hundred DR images from the public MESSIDOR database, manually labelled and prepared them for pixel segmentation, and tested the detection abilities of different model variants. We increased the diversity of the training sample by data augmentation, including tiling, flipping, and rotating the fundus images. The proposed approaches reached an acceptable mean average precision (mAP) in detecting DR lesions such as MA, HEMO, and EX, as well as a hallmark of the posterior pole of the eye, such as the optic disc. We compared our results with related works in the literature involving different neural networks. Our results are promising, but far from being ready for implementation into clinical practice. Accurate lesion detection is mandatory to ensure early and correct diagnoses. Future works will investigate lesion detection further, especially MA segmentation, with improved extraction techniques, image pre-processing, and standardized datasets.
Myopia is an eye disorder of global concern due to its increasing prevalence worldwide and its potential to cause sight-threatening conditions. Diagnosis is based on clinical tests such as objective cycloplegic refraction, distance visual acuity, and axial length measurements. Population-based screening is an early detection method that helps prevent uncorrected vision disorders. Advancements in technology and artificial intelligence (AI) applications in the medical field are improving the speed and efficiency of patient care programs. In an effort to provide a new, objective AI-based method for early myopia detection, we developed an algorithm based on the YOLOv8 convolutional neural network, capable of classifying eye fundus images from myopic and non-myopic patients. Preliminary results from an image set obtained from an Italian optometric practice show an overall accuracy of 85.00% and a precision and recall of 88.7% and 91.7%, respectively, in the internal validation dataset. This represents the beginning of a new paradigm, where AI is central to large screening programs aimed at preventing myopia and other avoidable blinding conditions and enabling early diagnosis and management.
This study evaluates the inter-device measurement properties of partial coherence interferometry (PCI) and spectral domain optical coherence tomography (SD-OCT) in measuring axial length, particularly for myopia management. We recruited 82 eyes from 41 adult participants with a mean age of 31.0 ± 17.6 years and a mean spherical equivalent of −2.20 ± 2.28 D. Axial length was measured using SD-OCT and PCI for both the right and left eyes. Agreement between the two measurements was assessed using Bland–Altman analysis, and graphs and values were compared with linear mixed models. The results show a near-to-zero and non-significant bias between measurements. The 95% limits of agreement showed a value of 0.06 mm. Both devices can accurately measure the axial length. OCT biometry performed with SD-OCT can be successfully interchanged with partial coherence interferometry, but they should be cautiously interchanged when performing longitudinal comparisons.
Myopia is a common refractive error with a rising prevalence worldwide, and its early diagnosis is crucial to prevent long-term visual impairment. This study presents an accessible, automated approach for detecting myopia from fundus photographs by analyzing the optic disc, using a deep learning model based on the YOLO (You Only Look Once) architecture, version 8 and 11. The pipeline was designed to be usable by eye care practitioners (ECPs) with no expertise in computer science. Fundus images were processed to extract the optic disc region using a custom-trained YOLOv8 model, and a subsequent classification algorithm determined the presence or absence of myopia based on features from the extracted region. The system was trained on a single-clinic dataset of 730 augmented images, with 98 images reserved for internal validation, and tested on 50 independent optic disc images. It achieved a high diagnostic accuracy, with strong sensitivity and F1 scores. Lightweight models such as YOLOv11-nano performed comparably to larger variants in the testing dataset (AUC 97.5% vs. 97.3%), effectively supporting myopia detection. This work highlights the feasibility of integrating AI-based screening tools into clinical practice without requiring advanced technical skills, offering a scalable and cost-effective solution to support early diagnosis of myopia in diverse healthcare settings.
Purpose: To evaluate the ability of computer vision models to detect myopia from standard colour fundus photographs and to compare their diagnostic performance with that of experienced ophthalmologists. Methods: A total of 324 retinal fundus images were labelled as myopic or non-myopic using cycloplegic subjective refraction. Images were acquired with a non-mydriatic 45° fundus camera and split into training, internal validation (80 images), and independent testing sets (50 images). YOLOv8 and YOLOv11 variants were trained for binary classification. Performance was evaluated using standard diagnostic metrics, with bootstrap confidence intervals and Holm–Bonferroni correction. Five ophthalmologists independently classified the test set, and their consensus was compared with model predictions using DeLong and McNemar tests. Results: On internal validation, YOLOv8-xl achieved the highest sensitivity (0.983) and MCC (0.646). YOLOv11 models showed greater variability, with the extra-large variant displaying degenerate behaviour. On the independent test set, YOLOv8-n achieved an AUC of 0.860 versus 0.753 for YOLOv11-s, with no significant difference (p = 0.174). Clinical consensus reached an AUC of 0.832 and did not differ from YOLOv8-n (p = 0.725). McNemar testing showed no difference between YOLOv8-n and clinicians (p = 1.000), while YOLOv11-s differed significantly (p < 0.001). Limitations include the small, class-imbalanced dataset and the limited number of ophthalmologists. Conclusions: YOLOv8 models can detect myopia from standard fundus photographs with performance comparable to experienced ophthalmologists, supporting their potential as complementary screening tools.
Objectives: The purpose of this study was to evaluate the ability of computer vision models to detect myopia from standard colour fundus photographs and to compare their diagnostic performance with that of experienced ophthalmologists. Methods: A previously published dataset of 324 retinal fundus images labelled as myopic or non-myopic based on cycloplegic refraction as used for model training and internal validation. Images were acquired using a non-mydriatic 45° fundus camera. Final model evaluation was performed on an independent test set of 50 images from different patients who were not included in the original dataset. YOLOv8 and YOLOv11 variants were trained for binary classification. Internal validation used patient-level cluster bootstrap confidence intervals, whereas image-level bootstrap confidence intervals were estimated for the independent test set. Pairwise model comparisons were adjusted using the Holm–Bonferroni correction. Five experienced ophthalmologists independently classified the test set, and their consensus was compared with the selected YOLO models using DeLong’s and exact McNemar tests. Results: Internal validation identified YOLOv8-m and YOLOv11-n as the best-performing models according to a predefined composite score used exclusively for model selection. On the independent test set, YOLOv11-n achieved the highest area under the curve (AUC = 0.889), followed by YOLOv8-m (0.806), although the difference was not statistically significant (DeLong test, p > 0.05). The clinical consensus achieved an AUC of 0.832, with no significant difference compared with either model. Exact McNemar testing likewise revealed no statistically significant differences in paired classification outcomes between either AI model and the clinical consensus. Limitations include the small, single-centre, class- and age-imbalanced dataset and the limited number of expert observers. Conclusions: Although neither YOLOv8-m nor YOLOv11-n showed statistically significant differences from the clinical consensus on this independent test set, these findings should be interpreted cautiously given the relatively small, single-centre study population. Larger multicentre studies with independent external validation are warranted to confirm the generalisability, robustness, and potential role of clinician-driven computer vision models as decision support tools for myopia screening.