記住我
REVIEW ARTICLE
Year : 2022 | Volume
: 29
| Issue : 1 | Page : 38-50
Multimodal imaging, tele-education, and telemedicine in retinopathy of prematurity
Nada H Almadhi1, Eliot R Dow2, RV Paul Chan3, Sulaiman M Alsulaiman1
1 Division of Vitreoretinal, King Khaled Eye Specialist Hospital, Riyadh, Saudi Arabia
2 Department of Ophthalmology, Jules Stein Eye Institute, University of California, Los Angeles, USA
3 Department of Ophthalmology and Visual Sciences, Illinois Eye and Ear Infirmary, University of Illinois, Chicago, Illinois, USA
Correspondence Address:
Dr. Sulaiman M Alsulaiman
Division of Vitreoretinal, King Khaled Eye Specialist Hospital, P.O. Box: 7191, Riyadh 11462
Saudi Arabia
Source of Support: None, Conflict of Interest: None
CheckDOI: 10.4103/meajo.meajo_56_22
Abstract
Retinopathy of prematurity (ROP) is a disease that affects retinal vasculature in premature infants and remains one of the leading causes of blindness in childhood worldwide. ROP screening can encounter some difficulties such as the lack of specialists and services in rural areas. The evolution of technology has helped address these issues and led to the emergence of state-of-the-art multimodal digital imaging devices such fundus cameras with its variable properties, optical coherence tomography (OCT), OCT angiography, and fluorescein angiography which has helped immensely in the process of improving ROP care and understanding the disease pathophysiology. Computer-based imaging analysis and deep learning have recently been demonstrating promising outcomes in regard to ROP diagnosis. Telemedicine is considered an acceptable alternative to clinical examination when optimal circumstances for ROP screening in certain areas are lacking, and the expansion of these programs has been reported. Tele-education programs in ROP have the potential to improve the quality of training to physicians to optimize ROP care.
Keywords: Imaging, retinopathy of prematurity, tele-education, telemedicine
Introduction 
Retinopathy of prematurity (ROP) is a vasoproliferative disease of premature infants and is one of the leading causes of childhood blindness worldwide.[1],[2],[3] It affects more than 65% of infants with birth weight <1251 g, and about 400–600 newborns become blind annually in the United States due to ROP.[4],[5],[6] Advances in neonatal care have led to the survival of lower birth-weight infants. ROP care is multidisciplinary and complex with fewer number of ophthalmologists willing to provide such care. Due to advances in fundus photography in infants and children, telemedicine ROP screening programs as well as tele-education are available or being developed.[7],[8],[9]
The first publication describing retinal imaging in a premature infant was in 1969.[10] They used a vertically mounted Carl Zeiss fundus camera with two operators to obtain the images.[10] Since then, a major advancement in pediatric fundus imaging was the introduction of RetCam in the 1990s. This has been instrumental in the diagnosis, documentation, and follow-up of pediatric vitreoretinal diseases.
In addition to photography, fundus fluorescein angiography (FA) may be a useful tool in ROP diagnosis. It has been used in the evaluation of retrolental fibroplasia in the late 1960s.[11],[12] Since then, several studies have described FA findings in ROP but have not established how these findings influence the management of ROP.[13],[14],[15],[16],[17],[18] More recently, FA was evaluated to assess its influence on zone and stage diagnosis and overall improvement in sensitivity and specificity of diagnosis.[19],[20]
The third edition of the International Committee for the Classification of ROP (ICROP3) was recently published with updated definitions. Nevertheless, the new classification retained the definition of the stages, which describes the disease severity and vascularity at the border of vascular and avascular retina. In addition, the 3 zones were still defined based on the location of the disease. However, posterior zone II was characterized as a zone that begins at the margin between zone I and II and extends for 2 disc diameters. It also maintained the definitions of plus and preplus disease but highlighted the fact that it is a continuous spectrum of disease severity. The aggressive ROP term was used to replace aggressive posterior ROP (APROP) to emphasize that the disease is not exclusively posterior.[21] Other changes included subcategorization of stage 5 ROP and notch definition.
In 2018, the American Academy of Pediatrics (AAP), American Academy of Ophthalmology (AAO), and American Association of Pediatric Ophthalmology and Strabismus (AAPOS) released an updated policy statement which acknowledged digital retinal imaging and remote interpretation as a tool for telemedicine screening since live examination by an experienced clinician might remain difficult to implement in some regions.[22] Nevertheless, numerous studies have shown imperfect inter-expert agreement in the diagnosis of ROP, especially plus disease, for both fundus imaging and binocular indirect ophthalmoscopy (BIO).[23],[24],[25],[26]
Computer-based image analysis (CBIA) systems in ROP screening may improve the accuracy of ROP diagnosis, and several semi-automated and fully automated systems have been developed.[24],[25],[27],[28],[29] These systems have not been integrated into clinical practice due to its limitations. The Imaging and Informatics in ROP “i-ROP” system were recently developed to address these limitations. It has the potential to perform comparably to experts in the diagnosis of plus disease.[30],[31] Convolutional neural network (CNN) is a type of artificial neural network that uses multiple filters and deep learning (DL) algorithms to serve higher-level information and has been recently demonstrating promising results in ROP screening and diagnosis.[32],[33],[34]
Optical coherence tomography (OCT) is a useful adjunct imaging modality in ROP screening which allows rapid acquisition of high-resolution images.[35] It has enabled clinicians to obtain OCT images in nonanesthetized, nonsedated neonates.[36] Commercially available portable OCT systems capable of image acquisition in the supine position include Envisue (Bioptigen, NC, USA), Spectralis (Heidelberg Engineering, Heidelberg, Germany) with Flex module, and Optovue (Freemont, CA) with different features for each.
OCT angiography (OCTA) is a recently introduced imaging modality that allows visualization of the retinal vasculature without dye injection.[37] Its utilization in ROP remains to be explored further.
This review article focuses on the imaging modalities used in ROP evaluation, including fundus photography, FA, OCT, and OCTA. It also discusses the use of images in telemedicine and tele-education. In addition, it highlights the potential use of postimaging computer-based analysis and artificial intelligence to enhance diagnostic accuracy.
Fundus Photography Contact Widefield imaging systems
RetCamTM (Natus Medical Inc)
RetCamTM III is the third version of RetCam system that allows for mydriatic color fundus photography as well as FA through a handheld camera with fiberoptic cable. Five interchangeable lenses (30°–130°) are available. A coupling agent is usually required to obtain good-quality images (640 × 480 pixels). Both anterior segment and posterior segment images and 2-min video capture can be obtained. Peripheral images to the ora serrata may require scleral indentation. The RetCamTM shuttle is a more affordable, compact, portable version of the RetCam. It has the same lens variety but lacks the capability of performing FA. The RetCam system has been utilized in telemedicine screening programs with success.[38],[39],[40],[41] The clarity of the retinal image is not always optimal as it may be affected by decreasing postmenstrual age (PMA), media opacity, anterior disease location, and darkly pigmented fundi.[42] Retcam Envision™, a new version of Retcam, has been released recently with a potentially improved illumination to enhance image quality.
PanocamTM (Visunex Medical Systems, Inc., Fremont, CA, USA)
The PanoCamTM Pro and PanoCamTM LT (Visunex Medical Systems) have been both approved in the United States for ophthalmic imaging in all newborn infants.[43] Along with its Cloud management system, PanoCam allows imaging through a wireless image-capturing unit. It has also an intelligent software that guides the photographer into capturing the recommended 6 standard fundus images. Two lenses are available (70° and 130° lenses) with 13 megapixel resolution. The integrated review software offers telemedicine capability. PanoCamTM LT is a lighter version that fits into a carrying case for easier transport and offers 130° field of view. The PanoCamTM, system does not currently have FA capability, but the modular design allows for upgradable diagnostic capabilities. It has not been enrolled in clinical trials yet and its utilization in research is pending.[44]
3nethra neoTM (Forus Health Pvt. Ltd. Bangalore, India)
The “3nethra Neo” designed in India by Forus Health, is an affordable, compact, nonwireless portable unibody imaging device. It provides 120° field of view and is telemedicine compatible, and the viewing software allows for image montaging. Neo device shows no significant ocular or systemic risks when used on awake preterm infants.[45] It is available in India and has the potential of international approval. In a pilot study, neo was found to be comparable to RetCam in detecting any stage ROP with good sensitivity and specificity.[46]
ICONTM (Phoenix Technology Group, Pleasanton, CA)
The icon is the latest commercially available pediatric retinal camera which has a different illumination technology that aims to deliver images of unparalleled contrast and resolution in darkly pigmented eyes. However, the field of view is 100° which is less than the previous cameras.
Prototype PedCam
Pedcam is an experimental pediatric camera that uses trans-pars-planar illumination for ultra-wide field fundus photography and enables a 200° field of view which allows visualization of the retina up to the ora serrata without the need of scleral depression. However, this needs an accurate identification of the pars plana since the mean width at full term is between 1.5 and 2 mm.[47],[48] This is a preliminary product that has been only tested on few ROP patients for proof of concept and is not yet available commercially.[47]
Noncontact ultra-widefield imaging
Optos (Dunfermline, Scotland, UK)
Optos camera provides a 200° noncontact image through an undilated pupil. It is especially valuable in evaluating vitreoretinal diseases in children due to the often limited examination and fast image acquisition.[49] Patel[50] and associates described the use of Optos to obtain images in premature neonates with ROP utilizing the “flying baby position,” a technique which may not be practical for routine clinical use. The advantages over RetCam include less exposure to bright light, which improves cooperation from photophobic patients, maintained image quality through mild media opacities and in pigmented fundi, wider field of view, and lack of contact avoiding undue pressure on the eye that may affect the accurate diagnosis of plus disease. However, limitations include lower image quality, visible eyelashes, poor detection of retinal detachment, and most importantly, the potential cardiac morbidity that could result from maneuvering the premature neonates to achieve the flying baby position.[50] The latest product of Optos, Silverstone, has an ultra-wide field swept-source OCT feature, which has a potential future use in ROP patients.
Fundus fluorescein angiography
FA provides important information in the assessment of pediatric vitreoretinal diseases such as Coats' disease, ocular tumors, familial exudative vitreoretinopathy, incontinentia pigmenti, sickle cell retinopathy, abusive head trauma, and others.[49],[51],[52],[53] Several studies focusing on ROP documented the safety and feasibility of oral as well as intravenous FA in premature infants.[12],[13],[14],[15],[16],[17],[18],[54]
Fluorescein angiography findings in retinopathy of prematurity
Zepeda-Romero et al. reported FA findings in premature infants as early as two weeks after birth, who eventually developed type 1 ROP. They found arteriovenous shunts surrounded by areas of capillary nonperfusion, rosary-bead-like hyperfluorescence along with tortuosity, and leakage from distal arterioles.[18] Lepore et al. described their findings in severe ROP requiring laser photocoagulation.[55] First, the arm-retina time was variable ranging from 4 to 53 s with delayed venular phase up to 2 min. This confirms the instability of retinal blood flow in premature infants. Choroidal circulation also showed variable filling. Second, findings at the vascular-avascular junction include leakage when stage 3 is present, abnormal vascular branching, circumferential vessels, and arteriovenous shunts (tangled or naked). Other findings include perivascular leakage from single or group of vessels, hyperfluorescent well-defined lesions corresponding to popcorns exceeding the degree observed on fundus images, capillary tuft formations, and focal capillary dilatations [Figure 1]. Third, features within the vascularized retina include areas of capillary dropout, absence of foveal avascular zone (FAZ), and diffuse hyperfluorescent macular lesions suggesting exudation.
Figure 1: Montage FA demonstrating leakage from extraretinal neovascularization in stage 3 ROP. FA: Fluorescein angiography, ROP: Retinopathy of prematurityEyes treated with anti-vascular endothelial growth factors
Anti-vascular endothelial growth factors (VEGF) therapy for ROP is frequently used as monotherapy for zone I, posterior zone II disease, and APROP.[56] Given the fact that the AAP discharge criteria for anti-VEGF-treated eyes require complete retinal vascularization and the known risk of late and very late (2.5–3 years of age) recurrence, including late traction retinal detachment, it is important to know the specific FA findings in these eyes.[57] Lepore et al. compared FA findings at 9 months in infants with type 1 ROP in zone I following laser photocoagulation or intravitreal bevacizumab (IVB). In the IVB-treated eyes, extensive areas of persistent avascular retina were observed along with capillary tufts, tangled arteriovenous shunts, and circumferential vessels. These changes were not seen in laser-treated eyes. Abnormalities within the vascularized retina were also seen more frequently in IVB-treated eyes, including capillary bed dropout (91.6% vs. 27.3%), absent FAZ, and posterior pole hyperfluorescent lesions (75.0% vs. 36.4%).[58] These findings were persistent up to 4 years of follow-up.[59]
Toy et al. found a characteristic, almost pathognomonic, regression pattern called ”scalloped regression” in IVB-treated eyes. This pattern is described as an irregular progression of leading vascular edge at terminal retinal vasculature.[60] Another study compared FA findings in infants treated with IVB for type 1 ROP to spontaneous regression. They found an unusual double-blunted vascular pattern mainly seen in the IVB group.[61] Interestingly, one study looked at five infants treated unilaterally with IVB for type 1 ROP in zone I or posterior zone II and compared retinal vascular growth to the fellow untreated, unaffected eyes. The result showed similar retinal vascular growth on FA in treated and untreated eyes.[62] Other anti-VEGF agents including ranibizumab and conbercept showed similar FA findings.[63],[64][Figure 2].
Figure 2: A montage FA showing peripheral vascular arrest with no abnormal vascular network and no leakage. FA: Fluorescein angiographyThe persistent avascular retina may be the drive for late reactivation. In Toy et al. study, the chronic vascular arrest occurred in 91% of eyes. The authors recommended laser ablation at 60 weeks' PMA in all eyes with persistent avascular retina.[60] More recently, FA was useful in determining four types of vascular regression patterns after IVB therapy, including complete vascularization, vascular arrest alone, vascular arrest with tortuosity, and reactivation. The majority (89%) did not reach full vascularization and 18% had ROP reactivation which significantly correlated to having a greater area of ischemia. The presence of tortuosity on FA could be a sign of persistent levels of VEGF and an early indicator of potential reactivation following IVB.[65]
Role of fluorescein angiography in retinopathy of prematurity diagnostic accuracy
It is clear that FA has enhanced the understanding of the disease process in ROP. However, its role in the management and diagnosis had not been illustrated until recently when Patel et al. examined the usefulness of FA in identifying the macula center to improve the accuracy of zone diagnosis. In the study, 32 sets of images (16 colored fundus images and 16 paired colored and FA images) were evaluated by nine ROP experts to identify the macular center and zone diagnosis. For each image set, a reference standard diagnosis (RSD) combining clinical examination and image-based diagnosis by three experts was established. In addition, a computer-facilitated diagnosis of the zone was established by measuring the distance from the disc to the macular center. Marginally significant improvement in sensitivity of zone diagnosis using colored images with FA (61.1%) compared to colored images alone (47%) (P = 0.07) was noted. However, there was no significant difference in macular center identification between colored images and FA.[20] The same group evaluated the influence of FA on the diagnosis and management of ROP. The nine experts were asked to provide a diagnosis and management plan for 32 sets of images similar to the design of the previous study. FA significantly improved the sensitivity of diagnosis of stage 3 or worse (P = 0.008), type 2 or worse (P = 0.013) and preplus or worse (P = 0.03). Moreover, FA improved inter-grader agreement for the diagnosis of ROP requiring treatment.[19] It seems that FA may add information to the examiner that enables a more accurate diagnosis.
Optical Coherence Tomography Angiography OCTA is a noninvasive imaging modality that uses motion contrast to obtain images of the retinal and choroidal vasculature at various levels.[37] Compared to FA, OCTA technology acquires three-dimensional scans in seconds, providing depth-resolved images in which the superficial capillary plexus, deep capillary plexus, and choriocapillaris can be visualized. The en face area is currently limited to 8 mm × 8 mm in the commercially available machines.[66] However, research prototypes have shown up to 60° of retina with good resolution.[67] Although it provides valuable information on lesion location and flow-void areas, leakage as well as low-flow vascular lesions cannot be seen. Furthermore, OCTA is subject to motion and projection artifacts, which limit the extracted information from the image. In addition, the current OCTA technology requires good fixation by the patient.[66],[68] Despite the difficulties in obtaining adequate imaging, OCTA use in pediatric retinal disease has been recently reported.
The first report by Vinekar et al. detected and monitored neovascularization complex regression using OCTA, utilizing the “flying baby technique,” of a preterm infant with APROP after receiving laser photoablation at 26 days of life. OCTA was repeated after 10 days of treatment and revealed reduction of vascular tortuosity and dilation with undetectable deeper vascular lesion flow.[69]
In 2017, Chen et al. reported using a swept source microscope integrated OCTA intraoperatively in 2 young patients with different pathologies. It was able to visualize small vasculature obscured by leakage on FA in ROP. However, the technology was limited by the narrow field of view.[70] Campbell et al.[71] used a prototype handheld ultra-wide-field OCT and OCTA device to obtain images in neonates with ROP in the neonatal intensive care unit (NICU) and operating room. The use of a noncontact method provided a 40° field of view and was obtained by directly holding the probe above the surface of the eye. Using a contact lens-based approach, both ultra-widefield (UWF) (approximately 100°×100°) OCT and 20°×20° (approximately 4 mm × 4 mm) OCTA scans in 2 s were obtained. The obtained OCTA images in one patient treated with laser demonstrated attenuated retinal flow in regions of previous laser treatment, loss of choriocapillaris with sparing of the larger choroidal vessels, and the absence of flow in the preretinal membranes. This theoretically suggests its potential benefit in providing an objective diagnosis and quantification of ROP staging and early detection of abnormal vascularization.[71]
As retinal microvascular structures in infants were exclusively assessed by histopathologic studies, in a recent study reported by Hsu et al., they were able to visualize superficial and deep vascular complex (SVC/DVC) at the perifoveal area and identify the boundaries of each complex in healthy full-term infants and using Spectralis flex module. The study also outlined and quantified the FAZ manually. Despite the comparable qualitative and quantitative measures of healthy retina in the evaluated age groups, a larger sample size of each age group is needed to establish a potential normative measure of retinal structures which can be used for future reference in ROP studies.[72] The same group utilized the same imaging system on 2 infants after treatment of type 1 ROP. The first case received bevacizumab in both eyes at 35 weeks' PMA and subsequent laser for the avascular retina, vitreoretinal surgery was performed in the right eye at 43 weeks' PMA for vitreous hemorrhage. The OCTA of the left eye at 73 weeks' PMA showed irregular SVC and large vessels diving into DVC.[73] The second case was imaged at 40 weeks' PMA after receiving laser photocoagulation in both eyes for zone II stage 3 ROP at 37 weeks' PMA. OCTA revealed a more posteriorly located SVC and lack of central flow in DVC. Although interesting, the nature of these changes after treatment cannot be ascertained.
Another study assessed the foveal microvascular characteristics in older children with a history of intravitreal ranibizumab (IVR) versus laser photocoagulation treatment for ROP using AngioPlex 5000°CTA. The central foveal vascular length density (VLD) and perfusion density (PD) were significantly lower in IVR group which may be explained by the decreased level of VEGF in children treated with IVR, resulting in decreased foveal vascular density.[74] In addition, FAZ was found to be smaller in laser photocoagulation group on OCTA. A plausible reason is that the destruction of avascular retina prevents peripheral migration and reorganization of the inner retinal cells, unlike IVR-treated eyes which allow inner retinal cell migration.[75]
Optical Coherence Tomography in the Diagnosis of Retinopathy of Prematurity Despite its widespread use in the diagnosis of other retinal diseases, OCT has played less of a role in the evaluation of ROP than other imaging modalities. Handheld imaging systems have overcome the challenge of performing OCT on premature neonates who cannot easily be imaged on upright devices. Despite the restricted view, OCT is useful in the detection of popcorn retinopathy, epiretinal membranes, cystoid macular changes, blood vessel abnormalities, retinoschisis and/or macular detachment all of which may have value in the treatment and prognosis of ROP.
Popcorn retinopathy refers to preretinal fibrovascular tissue that often lies posterior to the vascular ridge in ROP.[76] It has been observed in APROP lying near the optic nerve.[36] Other investigators found that popcorn retinopathy increases the risk of developing treatment-requiring ROP.[76]
Some investigators have also found subclinical epiretinal membranes (ERM) in nearly one-third of ROP patients using the spectral domain (SD) OCT, and when present, it deformed foveal architecture, suggesting significant structural consequences.[77] ERM was also observed more among ROP patients treated with laser photocoagulation.[78] Interestingly, in the study by Lee et al., ophthalmologists were unable to identify ERM by BIO in any of the cases in which they were detected with OCT.[77]
OCT also detected macular edema (ME) in half or more of infants with ROP stages 0-3.[79],[80] Another study looking specifically at premature neonates with stage 0-2 ROP found cystic ME in 16% of cases, whereas neonates with stage 0 or 1 had no cases of the pathology suggesting that ME may correlate with disease severity.[81] Other research has also found a stepwise increase in cystic ME by ROP stage.[82] Moreover, the edema, although transient and self-resolving, may affect visual acuity.[41] Altogether this suggests that cystic ME may be a useful marker to guide screening and treatment of ROP.
In addition, OCT may contribute to assessing the extent and nature of retinal detachment in ROP stages 4 and 5. Joshi et al. performed OCT imaging on infants with ROP stage 4A or 4B before pars plana vitrectomy and were able to appreciate abnormalities not detected by BIO. They surmised that the pathological findings identified before surgery may have predicted the variable outcomes following the procedure.[83]
OCT is not only helpful in assessing the foveal involvement but also in determining the nature of retinal elevation in stage 4 ROP. A study by Chen et al. looked into 19 eyes with adequate handheld spectral-domain OCT (SDOCT) imaging which managed to differentiate retinoschisis from the retinal detachment in areas of retinal elevation in eyes diagnosed as solely retinal detachment on clinical examination. Those who had retinoschisis without OCT evidence of retinal detachment may represent a new substage (stage 4-schisis) of stage 4 ROP, which may portend different prognosis and observation protocols.[84]
OCT was found to be useful in evaluating retinal vascular-avascular junction at the temporal peripheral retina. A study by Chen and others correlated the vascular-avascular junction SDOCT findings with previously reported histological features throughout the stages of ROP which were comparable. This offers an opportunity to observe the disease evolution over time in vivo and improve the understanding of retinal neovascular development in ROP. The same study monitored the regression of epiretinal neovascular tissue after 3 weeks of receiving IVB in infants with zone 1 stage 3 ROP. The investigators were able to visualize the development of 3 layered inner retina, the dramatic decrease in extraretinal vascular elevation, along with inner retinal split resolution.[85]
New developments in OCT imaging hold additional promise for ROP. Swept-source OCT employs a range of scanning wavelengths to reduce the imaging time by half compared to the SDOCT currently found in most clinics. Reducing the time required to keep the eye stationary by a few seconds is significant in working with young children.[86] Color Doppler OCT, which detects blood flow velocity, may also shed additional light on vascular pathology in ROP, especially for plus disease.[87] Although SDOCT produces decreased resolution at the outer layers of the retina compared to inner ones, enhanced depth imaging programs show greater clarity of the choroidal vasculature allowing investigators to explore the role of the choroid in ROP.[88]
Finally, tools to process OCT imaging will also contribute to its efficacy in diagnosing ROP. Algorithms for segmenting blood vessels may assist in the diagnosis of plus disease since the imaging modality, unlike BIO or photography, can provide detailed information about the three-dimensional position of vessels.[89]
[Figure 3] and [Figure 4] demonstrate the dependency on OCT as an adjunct modality to Retcam in diagnosing retinal traction.
Figure 3: (a) A preoperative left eye fundus photo with peripheral laser marks, contracting peripapillary tractional retinal detachment, epiretinal fibrosis at the macula, and overlying localized vitreous hemorrhage. (b) SDOCT capturing the area of traction. (c) A postoperative fundus image and SDOCT (d) after releasing the traction and removing fibrous tissues. SDOCT: Spectral-domain optical coherence tomography
Figure 4: (a) Fundus photo showing peripapillary retinal detachment, traction along the vascular arcades, and attached peripheral retina. (b) SDOCT shows dragged macula towards the optic disc with disrupted foveal depression. (c) Fundus photo 1-month postmembrane peeling, and dissection shows residual traction along the vascular arcades and over the disc, the macula is flat. (d) The foveal structure regained its normal contour after surgical Intervention. SDOCT: Spectral-domain optical coherence tomography Computer-Based Image Analysis for the Diagnosis of Plus Disease Because a significant weight of treatment decision depends on the presence of plus disease, its accurate diagnosis is extremely important.[90],[91] The level of vascular dilatation and tortuosity to make the diagnosis is based on a standard photograph with a narrow field of view of the posterior pole selected by expert consensus.[90] Several studies have confirmed a low degree of inter-grader agreement on plus disease diagnosis.[23],[24],[25],[26] Chiang and associates involved 22 ROP experts to interpret 34 sets of images of infants with ROP. Using three-level categorization (plus, preplus, or neither), all experts agreed on the same diagnosis in 4 of 34 images (12%), while in the 2-level categorization (plus or not plus), experts agreed in 7 of 34 images (21%).[23]
The potential of CBIA for plus disease diagnosis has been explored in the last two decades.[92] Software such as retinal image scale-space analysis, ROPtool, and CAIAR (computer-aided image analysis of the retina) have been demonstrated to be feasible in several studies.[24],[25],[27],[93],[94],[95],[96],[97]
There are several limitations of these programs. First, none of them is fully automated, meaning that an operator is required for proper functioning of the programs. Second, most studies have not addressed preplus disease. Third, most studies had imperfect RSD.
Imaging and Informatics in Retinopathy of Prematurity System To address the above-mentioned limitations, the “i-ROP” computer-based image analysis system was developed. It has several advantages over the above-mentioned image analysis systems. First, it was trained against a well-developed RSD derived from an image-based diagnosis made by three ROP experts combined with the diagnosis made during a clinical examination. Second, preplus disease was incorporated into the system. Third, this system could potentially utilize unprocessed images to diagnose preplus and plus disease. The first study evaluating the system's performance compared to RSD used manually segmented images. For a three-level classification (normal, plus, or preplus), the i-ROP system achieved 95% accuracy compared to the RSD. In comparison, the system was favorably comparable to individual accuracy of the 3 experts' grades.[31] Furthermore, using a computer-based algorithm that extracts vascular tortuosity and dilation features, the system outperformed 9 out of 11 experts.[30]
A more recent study implemented a trained CNN-DL system using a set of 5511 retinal images and 3-level diagnosis for plus disease. On an independent set of 100 images, the system achieved plus disease diagnosis with a 93% sensitivity and 94% specificity, which had comparable or better accuracy to expert human examiners. The algorithm achieved 100% sensitivity and 94% specificity in preplus diagnosis.[32]
In addition to 3-level plus disease diagnosis, the i-ROP consortium used an automated ROP vascular severity score to each image, scoring from 1 being normal retinal vasculature, to 9 which was a score for severe plus disease, stage 4 and 5 were excluded from the scoring. The study had 3 main key findings, first, the severity score scale demonstrates more quantifying measure for disease severity difference and helps in eliminating subjectivity and systematic bias between examiners which typically results in different management plan for all the components of ICROP by different physicians. Second, the ROP vascular severity score may potentially be useful in tracking disease progression. Third, the changes in quantitative ROP vascular severity score when longitudinally following images over time aid in identifying patients who ultimately need treatment for ROP.[98]
While APROP definition implies rapid progression, ICROP does not include the rate of progression in its diagnostic criteria. Hence, there has not been a formal method to evaluate its diagnostic features. Similar to the abovementioned study, using the i-ROP DL system, the group quantified the vascular severity score for each ICROP category and was found greatest in treatment requiring (TR) with AP-ROP 8.8 (interquartile range (IQR) 8.2–9.0), compared to 7.2 (IQR 5.3–8.7) for TR without AP-ROP (8.79 vs 7.19, P < 0.001).[99] Although graders tend to agree more on relative disease severity,[100] they found significant levels of inter-reader disagreement in diagnosing APROP (fair agreement range). Therefore, incorporating vascular severity score into diagnostic criteria for ROP while considering the clinical context may add some objectivity when determining the level of disease. When the severity score over time was analyzed, the rate of progression was found fastest in infants who developed APROP (P < 0.002 at 30–32 weeks). This suggests that many of the eyes that will progress to TR with APROP may be identified as early as 2 weeks before treatment, which may have implications on disease screening, risk modeling, and enabling earlier treatment threshold in future.[99]
To expand i-ROP DL system further, Redd et al. tested the ability of the system to identify clinically significant ROP and estimate the overall disease severity. Based on posterior pole images only and severity score of 3, the sensitivity and negative predictive value for type 1 ROP were 94% and 99.7%, respectively.[101] Two recent studies from China in which they developed deep neuronal networks for ROP, both systems performed well with high accuracy and their outcomes were comparable to grading experts.[33],[34]
Telemedicine Feasibility studies
Since first proposed in 1999, telemedicine ROP screening has been evaluated in multiple studies.[102] A joint technical report by AAP/AAO/AACO provided a summary that validated eight level-one-rated studies of telemedicine in the evaluation of ROP.[9] It compared independent masked readers to a reference standard ophthalmoscopic examination.[40],[103],[104],[105],[106],[107],[108] The largest study was the e-ROP group that was implemented in North America which included 5520 examinations from 1257 infants. A standard six-image set was obtained by a nonphysician staff and graded by two trained, nonphysician readers.[40] Eyes were considered for the presence of “referral warranted” ROP (RW-ROP), defined as (ROP in zone 1, the presence of plus disease, or the presence of stage 3 ROP at any time during the infant's hospital course). Findings in the e-ROP study were consistent with previous level one studies.[103],[104],[105],[106],[107],[108],[109] The authors of the joint technical report concluded that telemedicine does not supplant BIO for ROP screening. However, moderate-quality evidence supports the use of telemedicine to detect RWROP or “clinically significant” ROP (CSROP), a term that was also introduced and defined by the Photo-ROP group.[9],[105] In fact, the AAP in their latest policy statement in 2018 stated that “when implemented properly, telemedicine systems using wide-angle retinal images and clinical data may be used for preliminary ROP screening or as an adjunct to BIO for ROP screening.”[22]
Telemedicine “real-world” experience
Telemedicine for ROP has been utilized in several countries. In the US, one example of a successful ROP telemedicine program is the Stanford University Network for the Diagnosis of ROP. It uses telemedicine as the only screening tool for infants from six satellite NICUs in Northern California. All infants meeting the criteria for screening undergo wide-angle imaging with the RetCam by trained NICU nurses. Five or more images of each area of the eye are captured and then sent to a single ROP specialist. In case the images are not of good quality, the imaging session is repeated in 48 h. If images could not be obtained in the 2nd session, BIO is performed. BIO was also performed at least once to all infants.[110] In the 6-year results, of 1216 screened eyes, 22 infants required treatment, none developed adverse anatomical outcomes, and no cases requiring treatment were missed. The sensitivity, specificity, PPV, and NPV were 100%, 99.8%, 95.5%, and 100%, respectively.[111]
The Karnataka Internet Assisted Diagnosis of ROP (KIDROP) is a multicenter telemedicine program model that was initiated in 2008 to address the lack of ROP specialists in rural India. Images were obtained by trained technicians. Over a period of 4 years, the program reported the results of 7106 screened infants in which the ROP incidence was found to be 22.4% and TR-ROP was 3.6%. Another active mobile tele screening program called ROP eradication save our sight (SOS) was launched to serve unreached rural and sub-urban areas in India and aims to screen over 2000 babies per year. Since the initiation in 2015–2017, 8117 babies were screened and 127 were treated and prevented from going blind.[112]
In Saudi Arabia, a national ROP telemedicine program was launched in March 2019. The program aims to cover level-3 neonatal units across the country. With national ROP telemedicine screening guidelines in place, reporting software with the ability to obtain a second opinion, physician notification system (email and SMS), cameras distributed to NICUs, nurses training on photography, and network of trained retina specialist readers and treaters, the program currently covers 7 units.
Limitations of Telemedicine in Retinopathy of Prematurity Some studies evaluated the potential limitations of telemedicine. In a study by Quinn et al., they analyzed the discrepant RW-ROP findings between trained nonphysician graders and clinical examination from the previous e-ROP study using a consensus review of 4 ROP experts.[113] The study highlighted the limitations of both clinical examination and nonphysician graders in determining zone I, stage 3 and plus disease. The subtle nature of flat neovascularization in zone I and suboptimal image quality in peripheral pathology with 2-dimensional view make stage 3 diagnosis challenging for nonphysician graders. However, trained graders were more likely to correctly diagnose plus disease compared to clinical examination.
Another study highlighted inconsistent sensitivity of telemedical diagnosis of stage 4 and 5 ROP which raises the question whether telemedicine would accurately detect the major advanced morphological changes of ROP.[114]
Image quality can sometimes be affected by media haze, small pupils, or darkly pigmented fundi. This makes remote interpretation challenging. Therefore, a backup examiner must be available in case of ambiguity. Ungradable images were noted in few studies ranging between 8% and 11%.[103],[115] Furthermore, it is not possible with the current image quality to be able to discharge patients from acute phase screening without a confirmatory BIO. As per the joint technical report by AAP/AAO/AACO, BIO must be performed on the hospitalized infant when suitable for termination of acute phase evaluation before discharge from NICU, or within 72 h of the last remote digital fundus imaging examination, whichever is earlier. An outpatient follow-up examination is arranged with a qualified examiner to those discharged before meeting the termination criteria. The time of the exam must be arranged in accordance with ROP examination guideline requirement.[9]
Tele-Education for Retinopathy of Prematurity Many reports suggested a lack of sufficient exposure to ROP training for trainees.[116],[117],[118],[119],[120],[121] The emergence of tele-education programs in many countries has helped in improving ROP training. In the US and Canada, investigators have developed the Global Education Network for ROP tele-education system in collaboration with the imaging and informatics (i-ROP). The online system is comprised of a pretest and posttest, tutorial, and five modules each consisting of 5 clinical cases. Altogether, participants encounter bilateral color fundoscopic imaging from 36 infants with ROP representing all zones, stages, and categories of disease.[8] A pilot study involving residents from five programs in the United States and Canada found that the tele-education curriculum led to an increase in diagnostic accuracy across all classifications of ROP. This included a significant increase in sensitivity, particularly for APROP, and increase in specificity for stage 2 or worse ROP and preplus or worse disease.[8]
Another study involving ophthalmology residents in three middle-income countries also found similar benefits from the tele-education system.[7] In fact, following completion of the tele-education program, the participants' rates of sensitivity in detecting moderate, mild-or-worse, and TR-ROP exceeded the baseline rates achieved by pediatric ophthalmology fellows trained in the US programs.[118] Across both studies, a significant majority of participants rated themselves as having an adequate understanding of the diagnosis of ROP as a result of the tele-education system.[7],[8] Although additional testing remains to be done, the tele-education program appears to provide a scalable solution to the lack of ROP training encountered during ophthalmology residency and the dearth of available ROP educators worldwide.
Interestingly, trainees may also benefit from digital image-based instruction on the treatment of ROP. Four vitreoretinal fellows performed laser photocoagulation for ROP. Results from subsequent digital imaging of the retina were reviewed with the fellows to identify areas that they failed to treat. Skip lesions were present in 59% of eyes and follow-up treatment was performed in those areas.[122] This point-of-care feedback provides a powerful means to educate trainees on the treatment of ROP and is highly amenable to a t
留言 (0)