Automatic Refractive Error Estimation Using Deep Learning-Based Analysis of Red Reflex Images

Purpose/Background: We evaluate how a deep learning model can be applied to extract refractive error metrics from pupillary red reflex images taken by a low-cost handheld fundus camera. This could potentially provide a rapid and economical vision-screening method, allowing for early intervention to prevent myopic progression and reduce the socioeconomic burden associated with vision impairment in the later stages of life. Methods: Infrared and color images of pupillary crescents were extracted from eccentric photorefraction images of participants from Choithram Hospital in India and Dargaville Medical Center in New Zealand. The pre-processed images were then used to train different convolutional neural networks to predict refractive error in terms of spherical power and cylindrical power metrics. Results: The best-performing trained model achieved an overall accuracy of 75% for predicting spherical power using infrared images and a multiclass classifier. Conclusions: Even though the model's performance is not superior, the proposed method showed good usability of using red reflex images in estimating refractive error. Such an approach has never been experimented with before and can help guide researchers, especially when the future of eye care is moving towards highly portable and smartphone-based devices.

Reliability of Smart Phone Photographs for School Eye Screening

Smartphone photographs capturing Bruckner’s reflex have demonstrated reliability in identifying amblyogenic conditions in children. Assessing visual acuity for screening has been the traditional method since the inception of school screening. The present study aims to assess the reliability of smartphone photographs in detecting ocular morbidities in school children and to compare it with traditional vision screening. Two thousand five hundred and twenty school children underwent vision screening and smartphone cameraphotography by a trained research assistant followed by a comprehensive eye examination of all children by an ophthalmologist. Children with unaided visual acuity less than 6/12 in either of the eyes were graded as abnormal. Based upon the characteristics of the Bruckner’s reflex, the photographs were graded as normal or abnormal by two investigators blinded to the clinical findings. Statistical analysis was performed to compare the sensitivity and specificity of traditional vision screening and photograph based screening, considering comprehensive eye examination as the gold standard. The sensitivity and specificity of vision screening was 81.88% and 97.35% whereas for photographs it was 94.69% and 98.85% respectively. When the two methods were compared, the p value was <0.05. We conclude that smartphone photography is better than traditional vision screening for detecting ocular morbidities in school children.

Comparison between wide-field digital imaging system and the red reflex test for universal newborn eye screening in Brazil

PURPOSE

To compare neonatal eye screening using the red reflex test (RRT) versus the wide-field digital imaging (WFDI) system.

METHODS

Prospective cohort study. Newborns (n = 380, 760 eyes) in the Maternity Ward of Irmandade Santa Casa de Misericórdia de São Paulo hospital from May to July 2014 underwent RRT by a paediatrician and WFDI performed by the authors. Wide-field digital imaging (WFDI) images were analysed by the authors. Validity of the paediatrician's RRT was assessed by unweighted kappa [κ] statistic, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV).

RESULTS

While WFDI showed abnormalities in 130 eyes (17.1%), RRT was only abnormal in 13 eyes (1.7%). Wide-field digital imaging (WFDI) detected treatable retina pathology that RRT missed including hyphema, CMV retinitis, FEVR and a vitreous haemorrhage. The sensitivity of the paediatrician's RRT to detect abnormalities was poor at 0.77% (95% confidence interval, CI, 0.02%-4.21%) with a PPV of only 7.69% (95% CI, 1.08%-38.85%). Overall, there was no agreement between screening modalities (κ = -0.02, 95% CI, -0.05 to 0.01). The number needed to screen to detect ocular abnormalities using WFDI was 5.9 newborns and to detect treatable abnormalities was 76 newborns.

CONCLUSION

While RRT detects gross abnormalities that preclude visualization of the retina (i.e. media opacities and very large tumours), only WFDI consistently detects subtle treatable retina and optic nerve pathology. With a higher sensitivity than the current gold standard, universal WFDI allows for early detection and management of potentially blinding ophthalmic disease missed by RRT.

Beyond photography: evaluation of the consumer digital camera to identify strabismus and anisometropia by analyzing the Bruckner's reflex

Amblyopia screening is often either costly or laborious. We evaluated the Canon Powershot TX1 (CPTX1) digital camera as an efficient screener for amblyogenic risk factors (ARF). We included 138 subjects: 84-amblyopes and 54-normal. With the red-eye-reduction feature off, we obtained Bruckner reflex photographs of different sized crescents which suggested anisometropia, while asymmetrical brightness indicated strabismus; symmetry implied normalcy. Eight sets of randomly arranged 138 photographs were made. After training, 8 personnel, marked each as normal or abnormal. Of the 84 amblyopes, 42 were strabismus alone (SA), 36 had anisometropia alone (AA) while six were mixed amblyopes (MA). Overall mean sensitivity for amblyopes was 0.86 (95% CI: 0.83-0.89) and specificity 0.85 (95% CI: 0.77-0.93). Sub-group analyses on SA, AA and MA returned sensitivities of 0.86, 0.89 and 0.69, while specificities were 0.85 for all three. Overall Cohen's Kappa was 0.66 (95% CI: 0.62-0.71). The CPTX1 appears to be a feasible option to screen for ARF, although results need to be validated on appropriate age groups.