Colour slides or digital photography in diabetes screening - a comparison

A O S 2000

Clinial Science Colour slides or digital photography in diabetes screening – a comparison Marianne Henricsson, Cathrin Karlsson, Leif Ekholm, Pirjo Kaikkonen, Anders Sellman, Eva Steffert and Maria Tyrberg Department of Ophthalmology, Helsingborg Hospital, Helsingborg, Sweden

ABSTRACT. Purpose: The aim of this study was to compare digital images with slides in detecting and grading diabetic retinopathy, and to assess the retinopathy screening performed by ophthalmic nurses. Method: 283 consecutive patients were examined using digital colour and redfree photography (Topcon Imagenet System 1.53) and 35 mm slides (Topcon TRC-50 VT fundus camera, Kodachrome 64 colour film). The images were graded by the worst eye according to the Wisconsin classification by an ophthalmologist and ophthalmic nurse independently. Results: There was exact agreement between grades obtained from the colour slides and the digital colour images in 82% (weighted kappa 0.88; 95% CI 0.80– 0.96), and in 85% when redfree images were used as an adjunct to the digital colour images. There was a tendency towards undergrading of the digital colour images and overgrading of the digital redfree images, compared with the colour slides. Inter- and intragrader agreement (weighted kappa) varied between 0.77 and 0.84 for digital photography and between 0.88 and 0.90 for colour slides. Conclusion: Good to excellent agreement was found between the grading of colour slides and digital colour images, the latter, however, associated with a slightly lower reliability. The adjunct of redfree images seemed to facilitate the detection of retinopathic lesions. Key words: diabetic retinopathy – screening – digital images – colour slides. Acta Ophthalmol. Scand. 2000: 78: 164–168 Copyright c Acta Ophthalmol Scand 2000. ISSN 1395-3907

D

iabetic retinopathy is still the most common cause of blindness in the working age groups (Kahn & Hillier 1974; Evans et al. 1996). In the UK, 50% of diabetic patients registered as blind had had no previous eye examination (Clark et al. 1994). A primary objective of the St. Vincent declaration was to reduce new cases of blindness due to diabetes by one-third or more (Anonymous 1990). Screening for retinopathy before visual loss has occurred is essential in achieving this objective, since in most cases retinopathy can be treated successfully by photocoagulation (DRS Report

164

Number 8 1981; ETDRS Report Number 2 1987). Accordingly, screening programmes using fundus photography have been introduced in many countries, and in Sweden eye examinations are recommended at diagnosis and then every second year (SBU 1993). New diagnostic criteria for diabetes (Alberti & Zimmet 1998) will increase the number of diabetic patients by 7–8% (Groop et al. 1996), and consequently increase the demand for eye examinations. It is therefore important to develop inexpensive and simple methods to screen for retinopathy. Digital retinal cameras are capable of immediately dis-

playing a retinal image onto a high resolution computer monitor. A trained photographer can inform the patient at once about the retinal status. The objective of this study was to compare colour and redfree digital photography with 35 mm colour slides in detecting and grading diabetic retinopathy, and also to assess the retinopathy grading performed by ophthalmic nurses.

Patients A photographic screening programme for detection and control of retinopathy using 35 mm colour slides has been used for 10 years in Helsingborg, Sweden. It is estimated that 75% of the diabetic population is included in the programme (Henricsson et al. 1996). The patients are examined when diagnosed and then asked to return every second year. Patients with diagnosed retinopathy are followed up once a year. 283 consecutive diabetic patients from the screening programme were examined using both 35-mm colour slides and digital colour photography. Out of these, the last 80 patients were examined with digital monochrome, black-and-white images, as an adjunct to the digital colour images (redfree imaging). The retinopathy level of each patient was assigned the retinopathy level of the worse eye. Eyes with hazy media or eyes with branch or central venous occlusion were given the same grading level as the fellow eye. Four patients were excluded since both eyes were ungradable for causes other than diabetes. 279 patients were included in the study, 200 in the first and 79 in the second group, in which redfree photography was also used (Table 1).

A O S 2000 Table 1. Clinical characteristics of patients. Patients nΩ279 Age (years) Sex Men Women

59 (10–84) 167 (60) 112 (40)

Degree of retinopathy according to colour slides (worse eye), n (%) None (level 10) 142 (51) Very mild NPDR (level 21) 38 (14) Mild NPDR (level 31) 46 (16) Moderate NPDR (level 41) 30 (11) Severe NPDR (level 51) 3 (1) Panretinal lasertreated PDR 14 (5) (level 61) PDR (level 65) 4 (2) Not possible to grade 2 (1) because of DR (level 80) N (%), Median (range). Percentage do not add to 100 because of rounding.

Methods Colour slides Retinal photography was done in pharmacological mydriasis using an angle of 50æ with a Topcon TRC-50 VT fundus camera, (Tokyo Optical Co. Ltd, Japan) and Kodachrome 64 colour film (Kodak Ltd, England). The photographs covered fields 1–3 of the seven standard fields, with stereo pairs of the macula (field 2), and one field of the superior and one of the inferior fundus (standard fields 4–7). Digital colour photography Digital images were taken using the Topcon Imagenet System, version 1.53. The system has 640¿480 pixels resolution in true colour. A Sony DXC 930 video camera was used. The screen resolution was 800¿600 pixels in 24 bit colour (16.7 million colours) on a 42.5 cm high resolution Trinitron monitor. The 50æ digital images covered the same fields as the 35 mm slides (fields 1–3 and 2 fields of 4–7). Digital redfree black-and-white photography Digital redfree, monochrome, black-andwhite images were taken using the same Topcon Imagenet system, version 1.53. The green filter was the one supplied by the manufacturer. The system has 8 bits resolution with 256 grey scales. The 50æ images covered standard fields 1–3 and one field of the upper and one of the lower fundus.

Retinopathy grading A modification of the alternative classification of the Wisconsin Epidemiologic Study of Diabetic Retinopathy (Klein et al. 1986, 1989) and the Standard Photographs (SP) 2A, 8A and 10A (ETDRS Report Number 7 1991) were used to define the retinopathy level. This classification provides an overall retinopathy scale. Definitions: level 10 no diabetic retinopathy (DR) level 21 very mild DR, microaneurysms only, or blot haemorrhages, or cotton wool spots in the absence of microaneurysms level 31 mild non-proliferative diabetic retinopathy (NPDR), characterised mainly by hard exudates and/or mild retinal haemorrhages, less than those in SP 2A level 41 moderate NPDR, characterised by intraretinal haemorrhages, cotton wool spots and intraretinal microvascular abnormalities, but level 51 not met level 51 severe NPDR, characterised by one of the following: severe retinal haemorrhages; moderate intraretinal microvascular abnormalities equalling or exceeding those in SP 2A and 8A respectively, definite venous beading level 60 fibrous proliferations only level 61 scars of scatter photocoagulation, presumably directed at new vessels, but no evidence of levels 60 or 65 level 65 moderate proliferative diabetic retinopathy (PDR) characterised by definite new vessels on the disc, graded less than SP 10A, or moderate new vessels elsewhere or vitreous or preretinal haemorrhages, but level 70 not met level 70 PDR with high-risk characteristics (DRS Report Number 8 1981) level 80 grading impossible because of severe diabetic changes The presence of macular oedema was assessed as existing, non-existing or uncertain. Macular oedema was defined as retinal thickening (on colour slides only) and/or presence of hard exudates Østandard photograph 3 within one disc diameter of the centre of the macula (ETDRS Report Number 7 1991), and was considered present even when questionably so. Eyes that had earlier had laser treatment of the macula were not considered to have macular oedema if no current

hard exudates or retinal elevation could be detected. The presence of slight haze due to cataract was reported. Such haze may have prevented the detection of microaneurysms. The digital images and the slides were graded by the photographer (ophthalmic nurse) and by an ophthalmologist independently, at different time points, and in each group in random order. The intergrader variation was obtained by comparing these two gradings. The photographers and the ophthalmologists had no knowledge of the patients’ clinical data. Four ophthalmic nurses and three ophthalmologists participated in this study. Their experience with the retinopathy classification varied from 2 to 8 years. Colour slides were viewed against light boards using Donaldson’s stereo viewer (5¿ magnification). Digital images were graded with the hardware and software described above. The digital images were enhanced at the choice of the examiner to improve image quality by optimisation of contrast and brightness. Sharpening filters to improve visibility of lesions, and a digital zoom facility were used when required. The last 79 patients were examined using redfree black-and-white digital photography as an adjunct to the digital colour images. Firstly, retinopathy was graded on digital colour images and digital redfree images separately. Later, digital redfree images were graded together with the digital colour images. Using the latter approach, lesions found on the redfree images could be verified on the digital colour images. An independent adjudicator performed the final overall grading in eyes where the grades differed between the ophthalmologists and the ophthalmic nurses. The grader was very experienced, having used this grading system for eight years. After approximately 6 months, the ophthalmic nurses and the ophthalmologists regraded the digital colour images and the colour slides from the first 200 patients to obtain the intragrader reproducibility. The grading of the colour slides was regarded as the standard against which the new digital technique was compared.

Statistics The study was designed with a statistical power of 80% for detecting a difference in the grading of 10 retinopathy levels using digital colour photography and colour

165

A O S 2000 Table 2. Intra- and intergrader exact agreement.

Intragrader agreement ophthalmologist Intragrader agreement ophthalmic nurse Intergrader agreement ophthalmologist/ophthalmic nurse

Colour slides (weighted kappa statistic; 95% CI)

Digital colour images (weighted kappa statistic; 95% CI)

Digital redfree images (weighted kappa statistic; 95% CI)

0.83 (0.89; 0.79–0.98) 0.84 (0.88; 0.78–0.97) 0.85 (0.90; 0.82–0.98)

0.77 (0.84; 0.74–0.93) 0.70 (0.77; 0.68–0.87) 0.77 (0.84; 0.76–0.92)

0.71 (0.77; 0.63–0.91)

(CIΩconfidence interval).

slides, given a significance level of 0.05. Reliability was assessed using weighted kappa with weights equal to ªAbs(iªj)/ (1ªk), where i and j represent the ordinal categories for each measurement, and k represents the total number of categories (Fleiss 1981; Landis & Koch 1977). Eight categories were used, since only one patient was graded to level 60 (conferred to level 65), and none were graded to level 70.

Results Out of the 279 patients who were included in the study (Table 1), 14 were graded in only one eye; in 5 patients due to branch or central venous occlusion in one eye, and in 9 patients because of poor image quality due to cataract, which caused grading problems on both the slides and the digital images. Slight haze due to cataract was reported both on digital colour images and colour slides in ∂10% of the patients. The visibility of the redfree black-and-white images seemed to be better in the presence of mild cataract; haze was reported in that group in only ∂5%. Intragrader agreement The regrading of the first 200 patients showed an ophthalmologists intragrader agreement of 77% (weighted kappa 0.84; 95% CI 0.74–0.93) for digital colour images, and 83% (weighted kappa 0.89; 95% CI 0.79–0.98) agreement for colour slides. The corresponding figures for ophthalmic nurses were 70% (weighted kappa 0.77; 95% CI 0.68–0.87) and 84% (weighted kappa 0.88; 95% CI 0.78–0.97), respectively (Table 2). Intergrader agreement There was exact agreement between the ophthalmologists and the ophthalmic nurses in the grading of 8 retinopathy levels in 77% in digital colour images (weighted kappa 0.84; 95% CI 0.76–0.92),

166

and in 85% in colour slides (weighted kappa 0.90; 95% CI 0.82–0.98) (Table 2). Regarding digital redfree images, the intergrader agreement was 71% (weighted kappa 0.77; 95% CI 0.63–0.91).

Given that the retinopathy grade obtained from the colour slides was the standard, the sensitivity for detection of any retinopathy on digital images was 93%, the specificity was 91%, the positive predictive value 91%, and the negative predictive value 93% (Table 3). Macular oedema was diagnosed in 10 patients (3.5%) on digital colour images, and in 2 more patients (12 patients, 4%) on colour slides.

Digital colour images versus colour slides After clarification of the grades where disagreement occurred between the ophthalmic nurses and the ophthalmologists, there was exact agreement between grades obtained from the colour slides and the digital colour images in 82% (weighted kappa 0.88; 95% CI 0.80–0.96) (Table 3). The weighted kappa varied between the different categories and was highest in levels 10 and 61, 0.82 and 0.93 respectively, lowest in levels 21 and 41, 0.56 and 0.68 respectively, and intermediate, 0.71, in level 31. The levels 51, 65 and 80 had too few patients to assess agreement or kappa. Numbers above the diagonal line represent relative undergrading of the digital images in comparison to the colour slides, while numbers below the line represent overgrading. An undergrading was found mainly on the digital images, in 34 patients (12%) a lower level was found on the digital images than on the colour slides, and in 15 patients (5%) a higher level was given on the digital images than on the colour slides.

Digital redfree black-and-white images versus colour slides After clarification of the grades where disagreement occurred between the ophthalmic nurses and the ophthalmologists, there was a 66% exact agreement between grades obtained from the digital redfree black-and-white images and colour slides (weighted kappa 0.79; 95% CI 0.64–0.91) (Table 4). The level obtained from the digital redfree images was higher than on the colour slides in 25 patients (32%), and lower in 2 patients (3%). Digital colour and redfree black-andwhite images versus colour slides After clarification of the grades where disagreement occurred between the ophthalmic nurses and the ophthalmologists, there was an 85% exact agreement be-

Table 3. Retinopathy grades by grading method (digital colour images and colour slides). Colour slides

Digital colour images

Total

10 21 31 41 51 61 65 80

10

21

31

41

129 11 2

10 27 1

9 37

1 11 17 1

142

38

46

30

51

1 2

3

61

1 13

14

65

80

Total

2

139 48 51 18 4 14 3 2

2

279

1 3

4

A O S 2000 Table 4. Retinopathy grades by grading method (digital redfree images and colour slides). Colour slides

Digital redfree images

10 21 31 41 51 61 65

Total

10

21

19 13 1

11 6

31

13 4

41

51

61

65

Total

1

19 24 22 7 1 5 1

1

79

2 3 1 5

33

17

17

6

5

Table 5. Retinopathy grades by grading method (digital colour and redfree images and colour slides). Colour slides

Digital colour and redfree images

10 21 31 41 61 65

Total

tween grades obtained from the digital colour and redfree black-and-white images graded together, and colour slides (weighted kappa 0.86; 95% CI 0.71–1.00) (Table 5). The level obtained from the digital colour and redfree images was higher than on the colour slides in 9 patients (11%), and lower in 3 patients (5%).

Discussion Close agreement (82%; weighted kappa 0.88) was shown in the grading of 35 mm colour slides and digital colour images (Table 3). Weighted kappa statistic above 0.81 is considered as very good agreement (Landis & Koch 1977). Prior results have reported good to excellent agreement between retinopathy grades using a digital retinal imaging system compared to 35 mm colour slides (George et al. 1998). However, the resolution of the digital systems is still relatively low, and we found the digital images more difficult to grade. In fact, the inter- and intragrader reproducibility in the grading of digital images was lower compared with 35 mm

10

21

26 6 1

16 1

31

16 1

41

61

65

Total

1

26 22 21 4 5 1

1

79

3 3 5

33

17

17

6

5

colour slides (70–77% versus 83–85%) (Table 2). There was also a tendency towards undergrading of the digital colour images in comparison with colour slides. The poorest agreement was registered about level 21, which is characterised predominantly by the presence of microaneurysms. Level 61, panretinal lasertreated retinopathy, demonstrated the highest reproducibility. The system we used has a low image resolution (640¿480 pixels) which made the images fairly granular. This might explain why there was uncertainty about the microaneurysms. When digital redfree images were compared with colour slides, significantly more lesions were found on the redfree images than on the slides (Table 4), indicating an overgrading of the digital redfree images. Pigmentary spots and artefacts may have been misinterpreted as retinopathic lesions. The artefacts were mainly ascribed to dirt or dust in the optical system. The combined level of digital colour and redfree images versus colour slides showed a total agreement of 85%, with a tendency towards an overg-

rading of the digital colour and redfree black-and-white images (Table 5). Since the graders could check that the ‘‘spots’’ represented retinopathic lesions and not artefacts or pigmentary spots, this indicates that more lesions are detected on redfree black-and-white images than on 35 mm colour slides. The graders felt that they could detect microaneurysms and intraretinal microvascular abnormalities (IRMAs) more easily on redfree blackand-white images than on digital colour images or 35 mm colour slides. Media opacities seemed to be less disturbing in the grading of redfree black-and-white images than in the grading of digital colour photographs. Few patients in the study group had macular oedema, and the detection rate was similar using 35 mm colour slides and in digital colour images. This indicates that macular oedema was diagnosed predominantly by the presence of centrally located hard exudates also on colour slides. The image quality with electronic colour imaging did not seem to be quite as good as that of 35 mm photographic film, but more information could possibly be obtained with the adjunct of redfree black-and-white images. There are, however, other potential advantages in electronic imaging. The electronic image is available instantaneously so that the patient can be informed at once about the result, provided the photographer is an educated one. A patient sensitive to light could be photographed using weak background illumination. Furthermore, the instant availability of the electronic image confers the advantage that if there is any problem with quality, this is detected immediately so that a further photograph can be taken. The images can be readily stored on increasingly inexpensive and capacious computer storage media, and there is the possibility of electronic transfer of images. Redfree photography needs less flash intensity than the digital colour images, and the hardware demand for storage is substantially smaller. After the initial outlay on hardware and software, the electronic image itself is effectively without cost whereas the cost for colour film is substantial. It has been suggested that, in clinical practice, screening techniques should achieve a sensitivity of at least 80% and a specificity of 95% with no more than a 5% technical failure rate (Taylor 1996). If we consider the 5-field 50æ colour slides to be the standard, the digital colour

167

A O S 2000 photography fulfilled these criteria except for a slightly lower specificity (91%). It is likely that, with the adjunct of redfree black-and-white photography, more microaneurysms may be detected and some misclassifications may be avoided. The images were ungradable in both eyes in only 4 patients; in one eye in 14 patients (5%); i.e. a low technical failure rate. The high intergrader reproducibility between ophthalmic nurses and ophthalmologists suggests that trained and motivated ophthalmic nurses can provide an effective screening service for diabetic retinopathy. All patients can get instantaneous information about retinal status. In our practise, patients with no, very mild or mild retinopathy (levels 10–31), without indications of macular oedema, are taken care of by the nurses alone. This means taking and grading of images, information to the patient and to his or her general physician, and also planning for the next photography in 70– 80% of all diabetic patients. We conclude that digital photography by trained ophthalmic nurses provides a useful method for retinopathy screening. The digital system offers instant display, immediate access and improved storage of images. Improvement in technology will make the system even more attractive in the future.

Acknowledgements The study was supported by the Gorthon, Zoe´ga and Arvid Ohlsson Foundations, Helsingborg, Sweden.

168

References Alberti KGMM, Zimmet P for the WHO Consultation group (1998): Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: Diagnosis and classification of diabetes mellitus. Diabet Med 15: 539–553. Anonymous (1990): Diabetes care and research in Europe: The Saint Vincent Declaration. Diabet Med 7: 360. Clark JB, Grey RH, Lim KK & Burns-Cox CJ (1994): Loss of vision before ophthalmic referral in blind and partially sighted diabetics in Bristol. Br J Ophthalmol 78: 741–744. Diabetic Retinopathy Study Research Group (1981): Photocoagulation treatment of proliferative diabetic retinopathy: Clinical application of Diabetic Retinopathy Study (DRS) findings, DRS Report Number 8. Ophthalmology 88: 583–600. Early Treatment Diabetic Retinopathy Study Research Group (1987): Treatment techniques and clinical guidelines for photocoagulation of diabetic macular edema. ETDRS Report Number 2. Ophthalmology 94: 761–774. Early Treatment Diabetic Retinopathy Study Research Group (1991): Early Treatment Diabetic Retinopathy Study design and baseline patient characteristics. ETDRS Report Number 7. Ophthalmology 98 (suppl): 741–756. Evans J, Rooney C, Ashwood F, Dattani N & Wormald R (1996): Blindness and partial sight in England and Wales: April 1990– March 1991. Health Trends 28: 5–12. Fleiss J (1981): Statistical methods for rates and proportions. Wiley. George LD, Halliwell M, Hill R, Aldington SJ, Lusty J, Dunstan F & Owens DR (1998): A comparison of digital retinal images and 35 mm colour transparencies in detecting and grading diabetic retinopathy. Diabet Med 15: 250–253. Groop L, Forsblom C, Lehtovirta M, Tuomi T, Karanko S & Nisse´n M (1996): Metabolic consequences of a family history of

NIDDM: The Botnia Study. Evidence for sex-specific parental effects. Diabetes 45: 1585–1593. Henricsson M, Nilsson A, Groop L, Heijl A & Janzon L (1996): Prevalence of diabetic retinopathy in relation to age at onset of the diabetes, treatment, duration and glycaemic control. Acta Ophthalmol Scand 74: 523– 527. Kahn HA & Hillier R (1974): Blindness caused by diabetic retinopathy. Am J Ophthalmol 78: 58–67. Klein R, Klein BEK, Magli YL, Brothers RJ, Meuer SM & Davis MD (1986): An alternative method of grading diabetic retinopathy. Ophthalmology 93: 1183–1187. Klein R, Klein BEK, Moss SE, Davis MD & DeMets DL (1989): The Wisconsin Epidemiologic Study of Diabetic Retinopathy IX. Four-year incidence and progression of diabetic retinopathy when age at diagnosis is less than 30 years. Arch Ophthalmol 107: 237–243. Landis JR & Koch GG (1977): The measurement of observer agreement for categorial data. Biometrics 33: 159–174. Statens beredning fo¨r utva¨rdering av medicinsk metodik (SBU). Retinopati vid diabetes – va¨rdet av tidig uppta¨ckt. Stockholm: SBU 1993. Taylor R (1996): Practical community screening for diabetic retinopathy using the mobile retinal camera: report of a 12 centre study. Diabet Med 13: 946–952. Received on March 25th, 1999. Accepted on October 22nd, 1999.

Corresponding author: Dr. Marianne Henricsson Department of Ophthalmology Helsingborg Hospital S-25187 Helsingborg Sweden e-mail: m.henricsson/angelholm.mail.telia.com