Current Application of Digital Diagnosing Systems for Retinopathy of Prematurity

Highlights

• Retinopathy of prematurity is a blind-causing disease affecting children.

• Computer-based image analysis has been applied in the diagnosis of ROP.

• Automated diagnostic systems are advantageous and promising for ROP diagnosis.

Abstract

Background and Objective

Retinopathy of prematurity (ROP), a proliferative vascular eye disease, is one of the leading causes of blindness in childhood and prevails in premature infants with low-birth-weight. The recent progress in digital image analysis offers novel strategies for ROP diagnosis. This paper provides a comprehensive review on the development of digital diagnosing systems for ROP to software researchers. It may also be adopted as a guide to ophthalmologists for selecting the most suitable diagnostic software in the clinical setting, particularly for the remote ophthalmic support.

Methods

We review the latest literatures concerning the application of digital diagnosing systems for ROP. The diagnosing systems are analyzed and categorized. Articles published between 1998 and 2020 were screened with the two searching engines Pubmed and Google Scholar.

Results

Telemedicine is a method of remote image interpretation that can provide medical service to remote regions, and yet requires training to local operators. On the basis of image collection in telemedicine, computer-based image analytical systems for ROP were later developed. So far, the aforementioned systems have been mainly developed by virtue of classic machine learning, deep learning (DL) and multiple machine learning. During the past two decades, various computer-aided systems for ROP based on classic machine learning (e.g. RISA, ROPtool, CAIER) became available and have achieved satisfactory performance. Further, automated systems for ROP diagnosis based on DL are developed for clinical applications and exhibit high accuracy. Moreover, multiple instance learning is another method to establish an automated system for ROP detection besides DL, which, however, warrants further investigation in future.

Conclusion

At present, the incorporation of computer-based image analysis with telemedicine potentially enables the detection, supervision and in-time treatment of ROP for the preterm babies.

Introduction

Retinopathy of prematurity (ROP) is a vascular disease prevalent in premature infants, especially those with a birth weight lower than 1250 g. As a major cause of blindness in infants and young children [1], the incidence of ROP is rising due to the rapid improvement of rescue technology in neonatology worldwide. Nonetheless, the course of ROP is short, thereby making the effective therapeutic window relatively narrow. Once the disease progresses to the late stage or the tractional retinal detachment occurs, the prognosis of visual functions will be poor. Hence, diagnosing and treating ROP timely and accurately can effectively prevent blindness. According to the Early Treatment for Retinopathy of Prematurity (ETROP) trial [2], a timely clinical intervention is conducive to enhance the vision of high-risk patients with ROP, even though 9% became blind ultimately. Thus, the early screening and regular monitoring are particularly crucial. Plus disease in ROP is a key feature that requires treatment and can be identified by comparison with the published photographs regarding the arterial tortuosity and venous dilation of posterior retinal vessels. Therefore, conducting research on time- and labor-saving screening for ROP plus disease is necessary. Current therapies for ROP mainly include laser photocoagulation, intraocular injection, cryotherapy, scleral buckling and vitrectomy, based on the stages of the disease. Still, timely treatments are the key point for a good prognosis, given the fact that in the late stage of ROP, maneuvers like vitrectomy can manage to reduce a partial or complete anatomical impairment and yet fail to retain the normal visual function. Accordingly, it is critical to identify the treatment-warranted ROP (TW-ROP) in the high-risk infants.

In recent years, numerous morphological datasets have become accessible by digital images. By use of these large databases, retinal images can be analyzed efficiently with artificial intelligence (AI). Methods involving machine learning (ML) especially deep learning (DL) are capable of recognizing pathological characteristics in retinal diseases. Automated AI programs have been well developed and widely applied to screen diabetes in many countries. A breakthrough is that the world's first market-oriented AI medical device developed by Abramoff et al. was approved by the United States Food and Drug Administration (US FDA) in the April, 2018, which allowed automatic screening and diagnosis of diabetic retinopathy (DR) in diabetic adults using the autonomous DR and DME detection software (IDx-DR) [3]. The progresses in computer-aided technology are also gradually changing the diagnostic method for ROP. Although the traditional binocular indirect ophthalmoscopy (BIO) remains the golden standard for ROP screening, it requests the direct practice by experienced clinicians. The wide-angle digital retinal imaging system (Retcam) is increasingly utilized for fundus examination of ROP nowadays. Compared with BIO, Retcam is easier to operate with a shorter learning cycle, and in the meantime can be used for photographing, storage, output and transmission of multi-directional fundus images. Further, Retcam system is more conducive for follow-up, consultation, scientific research and education. These advantages of Retcam guarantee satisfactory results when applying it to diagnostic digital image analysis on ROP.

Telemedicine is a method of long-distance image interpretation, which provides medical service to remote regions but requires training to the local operators. Developed from the concept of image collection and analysis in telemedicine, various systems of computer-based image analysis using classic ML have been invented with promising outcomes. These systems were developed based on a small number of images and have relatively low demand for web server. However, they require manual delineation of selected features, and thus systemic bias may result from handcrafted extraction. Automated systems for ROP diagnosis based on DL have been developed for clinical applications. These systems are more time- and labor-saving compared with those based on classic ML. Unfortunately, they still call for large datasets covering various ROP features and different severities, together with high demand for web server. Notably, for convolutional neural networks, innovative programs are expected to be designed based on different datasets with the population from distinct countries, so that the networks can be modified properly in various situations. Moreover, it is difficult to compare the existing systems in a uniform way owing to the lack of a standard strategy. Consequently, it is called upon to establish a program able to compare the existing ROP programs. Also, multiple instance learning (MIL) is anticipated to be applicable for automated diagnosis in future.

In this work, we have reviewed the current application of digital diagnosing systems for ROP by use of classic ML methods, MIL and DL. We compared these systems with analysis parameters, outcomes, performances, etc (Table 1). We found that to date, each currently-used system has its own merits and disadvantages. An incorporation of telemedicine and computer-based image analysis potentially enhances detection, supervision and in-time treatment of ROP for the preterm babies.