Development and international validation of custom-engineered and code-free deep-learning models for detection of plus disease in retinopathy of prematurity: a retrospective study.

BACKGROUND: Retinopathy of prematurity (ROP), a leading cause of childhood blindness, is diagnosed through interval screening by paediatric ophthalmologists. However, improved survival of premature neonates coupled with a scarcity of available experts has raised concerns about the sustainability of this approach. We aimed to develop bespoke and code-free deep learning-based classifiers for plus disease, a hallmark of ROP, in an ethnically diverse population in London, UK, and externally validate them in ethnically, geographically, and socioeconomically diverse populations in four countries and three continents. Code-free deep learning is not reliant on the availability of expertly trained data scientists, thus being of particular potential benefit for low resource health-care settings. METHODS: This retrospective cohort study used retinal images from 1370 neonates admitted to a neonatal unit at Homerton University Hospital NHS Foundation Trust, London, UK, between 2008 and 2018. Images were acquired using a Retcam Version 2 device (Natus Medical, Pleasanton, CA, USA) on all babies who were either born at less than 32 weeks gestational age or had a birthweight of less than 1501 g. Each images was graded by two junior ophthalmologists with disagreements adjudicated by a senior paediatric ophthalmologist. Bespoke and code-free deep learning models (CFDL) were developed for the discrimination of healthy, pre-plus disease, and plus disease. Performance was assessed internally on 200 images with the majority vote of three senior paediatric ophthalmologists as the reference standard. External validation was on 338 retinal images from four separate datasets from the USA, Brazil, and Egypt with images derived from Retcam and the 3nethra neo device (Forus Health, Bengaluru, India). FINDINGS: Of the 7414 retinal images in the original dataset, 6141 images were used in the final development dataset. For the discrimination of healthy versus pre-plus or plus disease, the bespoke model had an area under the curve (AUC) of 0·986 (95% CI 0·973-0·996) and the CFDL model had an AUC of 0·989 (0·979-0·997) on the internal test set. Both models generalised well to external validation test sets acquired using the Retcam for discriminating healthy from pre-plus or plus disease (bespoke range was 0·975-1·000 and CFDL range was 0·969-0·995). The CFDL model was inferior to the bespoke model on discriminating pre-plus disease from healthy or plus disease in the USA dataset (CFDL 0·808 [95% CI 0·671-0·909, bespoke 0·942 [0·892-0·982]], p=0·0070). Performance also reduced when tested on the 3nethra neo imaging device (CFDL 0·865 [0·742-0·965] and bespoke 0·891 [0·783-0·977]). INTERPRETATION: Both bespoke and CFDL models conferred similar performance to senior paediatric ophthalmologists for discriminating healthy retinal images from ones with features of pre-plus or plus disease; however, CFDL models might generalise less well when considering minority classes. Care should be taken when testing on data acquired using alternative imaging devices from that used for the development dataset. Our study justifies further validation of plus disease classifiers in ROP screening and supports a potential role for code-free approaches to help prevent blindness in vulnerable neonates. FUNDING: National Institute for Health Research Biomedical Research Centre based at Moorfields Eye Hospital NHS Foundation Trust and the University College London Institute of Ophthalmology. TRANSLATIONS: For the Portuguese and Arabic translations of the abstract see Supplementary Materials section.

The frequency of severe corneal opacities in premature infants.

PURPOSE: To study the frequency and possible causes of severe corneal opacities in premature infants undergoing retinopathy of prematurity (ROP) screening in the neonatal intensive care unit (NICU) at a single institution. METHODS: The medical records of all infants screened for ROP in the NICU between January 2015 and March 2019 were reviewed retrospectively. Criteria for screening were gestational age (GA) of <32 weeks or birth weight (BW) <1501 g. Characteristics of premature infants with severe corneal opacities were extracted from the record. RESULTS: A total of 445 premature infants were screened during the 51-month period. The prevalence of severe corneal opacities was 1.4% (6 infants). The median GA of the affected infants was 24.5 weeks (range, 23-32), and the mean BW was 624 g (range, 500-860 g). Two infants had lagophthalmos; their corneal opacity was unilateral. All the infants were under continuous positive airway pressure (CPAP) for a prolonged period. All 4 infants with bilateral corneal opacities required treatment for severe ROP-laser (n =1), bevacizumab injection (n = 1), or both (n = 2). CONCLUSIONS: Severe corneal opacities in premature infants are rare but potentially sight threatening, because they can hinder the retinal examination. Lagophthalmos-related exposure keratopathy is an obvious risk factor. CPAP therapy may be another risk factor leading to the development of severe corneal opacities. Identification of infants at risk and prompt commencement of lubricants is necessary to avoid long-term corneal opacities.

Reduced utility of serum IGF-1 levels in predicting retinopathy of prematurity reflects maternal ethnicity.

AIMS: To validate known risk factors and identify a threshold level for serum insulin-like growth factor 1 (IGF-1) in the development of severe retinopathy of prematurity (ROP) in an ethnically diverse population at a tertiary neonatal unit, 2011-2013. METHODS: A prospective cohort masked study was conducted. Serum IGF-1 levels at 31, 32 and 33 weeks were measured and risk factor data collected including gestational age (GA), birth weight (BW), absolute weight gain (AWG) and maternal ethnicity. The eventual ROP outcome was divided into two groups: minimal ROP (Stages 0 and 1) and severe ROP (Stage 2 or worse including Type 1 ROP). RESULTS: 36 patients were recruited: 14 had minimal ROP and 22 severe ROP. Significant differences between the groups were found in GA, BW, AWG and IGF-1 at 32 and 33 weeks. There was minimal rise in IGF-1 in Stage 2 patients and/or black patients (p=0.0013) between 32 and 33 weeks but no pragmatic threshold level of IGF-1 that could distinguish between minimal or severe ROP. CONCLUSIONS: There were significant differences in GA, BW, AWG and IGF-1 at 32 and 33 weeks between those babies with severe ROP and those with minimal ROP. However, there was no threshold level of IGF-1 at a time point between 31 and 33 weeks that can be used to exclude a large proportion of babies from screening. We also found ethnic differences in IGF-1 levels with infants born to black mothers having significantly lower IGF-1 levels at 32 and 33 weeks gestation. The determination of ROP risk using IGF-1 is a race-specific phenomenon.

Developing a model system for teaching goniotomy.

PURPOSE: To design a model-system instruction course to prepare trainees for performing goniotomy on patients. DESIGN: Experimental study. PARTICIPANTS: Three pediatric ophthalmology fellows and 1 recent graduate of the fellowship program. METHODS: We piloted 3 model systems: human cadaver eyes with and without Marty the Surgical Simulator (Iatrotech Inc., Del Mar, CA, hereafter referred to as Marty Head) eye supporting system and artificial eyes in Marty Head. For improving intraocular view in cadaver eyes, we used epithelial scrapping, intracameral viscoelastic, and/or intracameral lubricating jelly. Each trainee underwent a training course, including background reading, didactic lecture, and video review followed by goniotomy on model systems with the operating microscope. MAIN OUTCOME MEASURES: Each trainee evaluated each step of the course using modified 5-point Likert scales. Reading material and videos were evaluated for usefulness, readability/quality, and new information obtained. Each model system was evaluated for visibility, ease of setup, surgical feel, and transferability to live surgery. In the end, each trainee was asked to assess the overall course for usefulness. RESULTS: Trainees rated the reading materials and video highly for their usefulness and quality, yet they believed they did not learn substantial amounts of new information. Visibility and ease of setup was best with artificial eyes in the Marty Head model. Human cadaver eyes in the Marty Head provided somewhat less visibility and ease of setup, but the perceived feel and transferability to live surgery was slightly better than with artificial eyes. Cadaver eyes without the Marty Head got the lowest rating in all categories. At the end of the course, all participants felt more confident and ready to perform goniotomy on patients. All recommended this course as part of pediatric ophthalmology fellowship training programs. CONCLUSIONS: This model system instruction course can assist trainees in learning to perform goniotomy while potentially lessening the risks to patients.

Surgery for residual convergence excess esotropia.

The outcome of bilateral medial rectus posterior fixation sutures +/- central tenotomy was assessed as a secondary procedure for residual convergence excess esotropia in 11 patients. Ten had previously undergone bilateral medial rectus recessions. One had recess/resect surgery on the deviating eye. The average preoperative near angle was 30 prism diopters with a range of 16 to 45 prism diopters. Eight patients underwent bilateral medial rectus posterior fixation sutures with central tenotomy. Two had bilateral medial rectus posterior fixation sutures only, and one had bilateral medial rectus posterior fixation suture, a lateral rectus resection, and an inferior oblique disinsertion. The postoperative near angle ranged from 4-30 prism diopters, with mean of 12 prism diopters. Five patients demonstrated some stereopsis preoperatively, all needing bifocals. Postoperatively, nine patients demonstrated an improvement in stereopsis, none needing bifocals. Two showed smaller near angles and better control without bifocals. Final stereopsis ranged from 30 seconds of arc to 800 seconds of arc. We feel that bilateral medial rectus posterior fixation sutures with or without central tenotomy is a viable secondary procedure for residual convergence excess esotropia.