Fairness and generalisability in deep learning of retinopathy of prematurity screening algorithms: a literature review

BACKGROUND

Retinopathy of prematurity (ROP) is a vasoproliferative disease responsible for more than 30 000 blind children worldwide. Its diagnosis and treatment are challenging due to the lack of specialists, divergent diagnostic concordance and variation in classification standards. While artificial intelligence (AI) can address the shortage of professionals and provide more cost-effective management, its development needs fairness, generalisability and bias controls prior to deployment to avoid producing harmful unpredictable results. This review aims to compare AI and ROP study's characteristics, fairness and generalisability efforts.

METHODS

Our review yielded 220 articles, of which 18 were included after full-text assessment. The articles were classified into ROP severity grading, plus detection, detecting treatment requiring, ROP prediction and detection of retinal zones.

RESULTS

All the article's authors and included patients are from middle-income and high-income countries, with no low-income countries, South America, Australia and Africa Continents representation.Code is available in two articles and in one on request, while data are not available in any article. 88.9% of the studies use the same retinal camera. In two articles, patients' sex was described, but none applied a bias control in their models.

CONCLUSION

The reviewed articles included 180 228 images and reported good metrics, but fairness, generalisability and bias control remained limited. Reproducibility is also a critical limitation, with few articles sharing codes and none sharing data. Fair and generalisable ROP and AI studies are needed that include diverse datasets, data and code sharing, collaborative research, and bias control to avoid unpredictable and harmful deployments.

Intravitreal Angiostrongylus cantonensis: first case report in South America

This study reports the first case of intravitreal angiostrongyliasis in South America treated with posterior worm removal via pars plana vitrectomy. This was a retrospective, observational case study. Data from medical charts, wide-field digital imaging, ocular ultrasound, and visual evoked potential studies were reviewed. A 20-month-old boy presented with eosinophilic meningitis and right eye exotropia. Polymerase chain reaction analysis of the cerebrospinal fluid showed a positive result for Angiostrongylus cantonensis. Fundus examination revealed a pale optic disc, subretinal tracks, vitreous opacities, peripheral tractional retinal detachment, and a dead worm in the vitreous cavity. The patient underwent pars plana vitrectomy with worm removal. This case report illustrates the first case of intravitreal angiostrongyliasis in South America, possibly related to the uncontrolled spread of an exotic invasive species of snail.

Macular hole: 10 and 20-MHz ultrasound and spectral-domain optical coherence tomography

PURPOSE

Optical coherence tomography (OCT) is valuable for macula evaluation. However, as this technique relies on light energy it cannot be performed in the presence of opaque media. In such cases, the ultrasound (US) may predict some macular features. The aim of this study was to characterize images obtained by ultrasound with 10 and 20-MHz transducers comparing to OCT, as well as to analyze the relationship between the vitreous and retina in eyes with macular hole (MH).

METHODS

29 eyes of 22 patients with biomicroscopic evidence of MH at different stages were included. All patients were evaluated using ultrasonography with 10 and 20-MHz transducers and OCT.

RESULTS

OCT identified signs of MH in 25 of 29 eyes. The remaining 4 cases not identified by US were pseudoholes caused by epiretinal membranes. In MH stages I (2 eyes) and II (1 eye), both transducers were not useful to analyze the macular thickening, but suggestive findings as macular irregularity, operculum or partial posterior vitreous detachment (PVD) were highlighted. In stages III (14 eyes) and IV (5 eyes), both transducers identified the double hump irregularity and thickening. US could measure the macular thickness and other suggestive findings for MH: operculum, vitreomacular traction and partial or complete PVD. In cases of pseudoholes, US identified irregularities macular contour and a discrete depression.

CONCLUSION

10-MHz US was useful for an overall assessment of the vitreous body as well as its relationship to the retina. The 20-MHz transducer allowed valuable information on the vitreomacular interface and macular contour. OCT provides superior quality for fine morphological study of macular area, except in cases of opaque media. In these cases, and even if OCT is not available, the combined US study is able to provide a valid evaluation of the macular area improving therapeutic approach.

Standard full-field electroretinography in healthy preterm infants

The purpose of this study was to determine electroretinographic parameters according to the standard protocol from the International Society for Clinical Electrophysiology of Vision (ISCEV) in healthy preterm infants with normal fundus. Seventeen healthy preterm infants with normal fundus were recruited and divided in two age groups: 3-week group, nine infants with mean adjusted age at test = 2.67 +/- 0.92 weeks and 8-week group, eight infants with mean adjusted age at test = 7.92 +/- 1.72 weeks. Full-field ERGs were obtained with a Burian-Allen bipolar contact lens electrode from the anesthetized cornea in one eye, through a fully dilated pupil after 30 min of dark adaptation. The standard ISCEV protocol was used and the following responses were recorded: rod, maximal, oscillatory potentials, cone and 30 Hz flicker. Median values and 1st, 5th, 95th and 99th percentiles for amplitude and implicit time are described for both age groups. There was statistically larger amplitude for 30 Hz flicker (t = 2.191; p = 0.046) and for cone response (t = 2.307; p = 0.044) in the 8-week-old group. Statistically shorter implicit times were found in 8-week group for rod response (t = 3.219; p = 0.015), cone response (t = 2.839; p = 0.016) and flicker response (t = 3.326; p = 0.005). Shortening of implicit time was evident in the older group of preterms and this finding is consistent with other maturational studies confirming the anatomical and functional development of the photoreceptors. Medians and ranges between the 1st and 99th and the 5th and 95th percentiles can be used as a baseline for future comparisons with infants with ROP.