Effect of digital image compression on screening for diabetic retinopathy

Anterior ophthalmic imaging

Improvements in imaging chips and computer processing power have brought major advances in imaging of the anterior eye. Digitally captured images can be visualised immediately and can be stored and retrieved easily. Anterior ocular imaging techniques using slitlamp biomicroscopy, corneal topography, confocal microscopy, optical coherence tomography (OCT), ultrasonic biomicroscopy, computerised tomography (CT) and magnetic resonance imaging (MRI) are reviewed. Conventional photographic imaging can be used to quantify corneal topography, corneal thickness and transparency, anterior chamber depth and lateral angle and crystalline lens position, curvature, thickness and transparency. Additionally, the effects of tumours, foreign bodies and trauma can be localised, the corneal layers can be examined and the tear film thickness assessed.

On the quantitative effects of compression of retinal fundus images on morphometric vascular measurements in VAMPIRE

BACKGROUND AND OBJECTIVES

This paper reports a quantitative analysis of the effects of joint photographic experts group (JPEG) image compression of retinal fundus camera images on automatic vessel segmentation and on morphometric vascular measurements derived from it, including vessel width, tortuosity and fractal dimension.

METHODS

Measurements are computed with vascular assessment and measurement platform for images of the retina (VAMPIRE), a specialized software application adopted in many international studies on retinal biomarkers. For reproducibility, we use three public archives of fundus images (digital retinal images for vessel extraction (DRIVE), automated retinal image analyzer (ARIA), high-resolution fundus (HRF)). We generate compressed versions of original images in a range of representative levels.

RESULTS

We compare the resulting vessel segmentations with ground truth maps and morphological measurements of the vascular network with those obtained from the original (uncompressed) images. We assess the segmentation quality with sensitivity, specificity, accuracy, area under the curve and Dice coefficient. We assess the agreement between VAMPIRE measurements from compressed and uncompressed images with correlation, intra-class correlation and Bland-Altman analysis.

CONCLUSIONS

Results suggest that VAMPIRE width-related measurements (central retinal artery equivalent (CRAE), central retinal vein equivalent (CRVE), arteriolar-venular width ratio (AVR)), the fractal dimension (FD) and arteriolar tortuosity have excellent agreement with those from the original images, remaining substantially stable even for strong loss of quality (20% of the original), suggesting the suitability of VAMPIRE in association studies with compressed images.

Technical and imaging factors influencing performance of deep learning systems for diabetic retinopathy

Deep learning (DL) has been shown to be effective in developing diabetic retinopathy (DR) algorithms, possibly tackling financial and manpower challenges hindering implementation of DR screening. However, our systematic review of the literature reveals few studies studied the impact of different factors on these DL algorithms, that are important for clinical deployment in real-world settings. Using 455,491 retinal images, we evaluated two technical and three image-related factors in detection of referable DR. For technical factors, the performances of four DL models (VGGNet, ResNet, DenseNet, Ensemble) and two computational frameworks (Caffe, TensorFlow) were evaluated while for image-related factors, we evaluated image compression levels (reducing image size, 350, 300, 250, 200, 150 KB), number of fields (7-field, 2-field, 1-field) and media clarity (pseudophakic vs phakic). In detection of referable DR, four DL models showed comparable diagnostic performance (AUC 0.936-0.944). To develop the VGGNet model, two computational frameworks had similar AUC (0.936). The DL performance dropped when image size decreased below 250 KB (AUC 0.936, 0.900, p < 0.001). The DL performance performed better when there were increased number of fields (dataset 1: 2-field vs 1-field-AUC 0.936 vs 0.908, p < 0.001; dataset 2: 7-field vs 2-field vs 1-field, AUC 0.949 vs 0.911 vs 0.895). DL performed better in the pseudophakic than phakic eyes (AUC 0.918 vs 0.833, p < 0.001). Various image-related factors play more significant roles than technical factors in determining the diagnostic performance, suggesting the importance of having robust training and testing datasets for DL training and deployment in the real-world settings.

Retinal video recordings at different compression levels: a novel video-based imaging technology for diabetic retinopathy screening

BACKGROUND

To evaluate the optimal compression level of retinal color digital video recordings, a novel video-based imaging technology, in screening for diabetic retinopathy (DR).

DESIGN

Evaluation of a diagnostic technique.

METHODS

A total of 36 retinal videos, captured using EyeScan (Ophthalmic Imaging System), were compressed from original uncompressed file size of 1 GB (gigabyte) to four different compression levels-100 MB (megabyte) (Group 1); 30 MB (Group 2); 20 MB (Group 3); and 5 MB (Group 4). The videos were subsequently interpreted by an ophthalmologist and a resident using the International Clinical Diabetic Retinopathy Severity Scales.

MAIN OUTCOME MEASURES

The sensitivity, specificity and κ coefficient for DR grading detected by were calculated for each compression level (Groups 1-4), with reference to the original uncompressed retinal videos.

RESULTS

Groups 1, 2, and 3 graded by both readers had sensitivity and specificity >90% in detecting DR, whereas for group 4, the sensitivity and specificity were 70.6% and 94.7% for ophthalmologist and 80.0% and 72.2% medical officer, respectively. The κ correlation in detecting DR for groups 1, 2, and 3 were >0.95, whereas for Group 4, the κ was 0.76 and 0.66 for ophthalmologist and medical officer, respectively.

CONCLUSION

Retinal video recording is a novel and effective DR screening technique with high sensitivity, specificity and κ correlation. With its compressibility, this is a potential effective technique that can be widely implemented in a routine, mobile, and tele-ophthalmology setting for DR screening services.

Screening for diabetic retinopathy: review of current methods

OBJECTIVE

To review the current screening methods for diabetic retinopathy, with a focus on nonmydriatic digital fundus photography.

METHODS

Articles from Medline were reviewed from 1976 to November 2011 for different combinations of the words "diabetic retinopathy," "screening," "fundus photography," and "nonmydriasis."

RESULTS

Because of its ease of use and cost-effectiveness, digital fundus photography is becoming the preferred method for diagnosing diabetic retinopathy. Current research has proven that pupillary dilation is not a necessary step in the fundus examination, although it reduces the number of unnecessary referrals to ophthalmologists. Automated grading systems, while saving time and reducing human error, still need refinement before they can replace manual grading by trained ophthalmologists.

CONCLUSION

Nonmydriatic digital fundus photography with manual grading by a trained technician is an acceptable method of screening for diabetic retinopathy.

Grading diabetic retinopathy severity from compressed digital retinal images compared with uncompressed images and film

PURPOSE

To compare research grading of diabetic retinopathy (DR) severity level from compressed digital images versus uncompressed images and film.

METHODS

Compressed (JPEG2000, 37:1) digital images (C) were compared with uncompressed digital (U) and film (F) stereoscopic photographs from a 152-eye cohort with full-spectrum Early Treatment Diabetic Retinopathy Study severity levels for agreement on severity level, DR presence with ascending severity threshold, presence of DR index lesions, and repeatability of grading.

RESULTS

Classification of Early Treatment Diabetic Retinopathy Study severity levels from C images agreed substantially with results from F images (κ = 0.60, κ(w) [linear weighted] = 0.86) and uncompressed digital images (κ = 0.76, κ(w) = 0.92). For agreement of uncompressed digital versus F images, κ = 0.62 and κ(w) = 0.86. Distribution of Early Treatment Diabetic Retinopathy Study levels was not significantly different between C and F images (P = 0.09, Bhapkar's test for marginal homogeneity). For presence/absence of DR at 8 ascending severity thresholds, agreement between C and F was "almost perfect" (κ ≥ 0.8). Agreement on severity level between readers with C images was at least as good as that with uncompressed digital image or F. Repeatability of severity threshold grading between readers was similar using C or F images. For identifying individual DR lesions, agreement between C and F ranged from "moderate" to "perfect." Agreement of grading venous beading from C was slightly lower than from F.

CONCLUSION

Full Early Treatment Diabetic Retinopathy Study scale DR severity level grading using C images is comparable to that using U images or film.

Evaluation of the effect of JPEG and JPEG2000 image compression on the detection of diabetic retinopathy

AIMS

To compare the effect of classic Joint Photographic Experts Group (JPEG) and JPEG2000 compression algorithms on detection of diabetic retinopathy (DR) lesions.

METHODS

In total, 45 colour fundus photographs obtained with a digital nonmydriatic fundus camera were saved in uncompressed Tagged Interchanged Files Format (TIFF) format (1.26 MB). They were graded jointly by two retinal specialists at a 1 month interval for soft exudates, hard exudates, macular oedema, newvessels, intraretinal microvascular abnormalities (IRMA), and retinal haemorrhages and/or microaneurysms. They were compressed to 118, 58, 41, and 27 KB by both algorithms and 24 KB by classic JPEG, placed in random order and graded again jointly by the two retina specialists. Subjective image quality was graded, and sensitivity, specificity, positive and negative predictive values, and kappa statistic were calculated for all lesions at all compression ratios.

RESULTS

Compression to 118 KB showed no effect on image quality and kappa values were high (0.94-1). Image degradation became important at 27 KB for both algorithms. At high compression levels, IRMA and HMA detection were most affected with JPEG2000 performing slightly better than classic JPEG.

CONCLUSION

Performance of classic JPEG and JPEG2000 algorithms is equivalent when compressing digital images of DR lesions from 1.26 MB to 118 KB and 58 KB. Higher compression ratios show slightly better results with JPEG2000 compression, but may be insufficient for screening purposes.

Retinopathy screening in individuals with type 2 diabetes: who, how, how often, and at what cost—an epidemiologic review

BACKGROUND

Despite treatment regimens, which can reduce blindness, diabetic retinopathy remains the leading cause of irreversible blindness for persons less than 75 years. Screening for diabetic retinopathy in type 1 diabetes mellitus has proven to be both efficacious and cost effective. Although treatment has been shown to be equally efficacious for type 2 diabetes, the effectiveness and cost benefit of screening are less clear.

METHODS

A systematic review of the recent literature on the epidemiology of retinopathy in type 2 diabetics was carried out. Specific issues related to the changing epidemiology of diabetes and diabetic retinopathy, as well as the effectiveness, and cost benefit of screening were evaluated.

CONCLUSIONS

Diabetic retinopathy is at the convergence of several epidemiologic trends. Although the prevalence of diabetic retinopathy within the population of diagnosed diabetics may be decreasing because of aggressive risk factor reduction, any gains achieved are likely to be negated by the large increase in the prevalence of type 2 diabetes in the population owing to obesity. The epidemiology findings suggest that certain persons with well-controlled type 2 diabetes may not warrant yearly examination for diabetic retinopathy; however, given the low percentage of diabetics who are currently screened for retinopathy, it would seem unwise to recommend changes in the timing of examination. Evidence appears to suggest that photoscreening is a viable alternative to a screening examination.

Screening for diabetic retinopathy by one-field, non-mydriatic, 45 degrees digital photography is inadequate

PURPOSE

To evaluate the sensitivity and specificity of one-field, non-mydriatic, 45 degrees digital photography for screening for diabetic retinopathy compared to indirect ophthalmoscopy using a slit-lamp, the reference standard.

METHODS

A total of 100 consecutive diabetic patients (200 eyes) who underwent digital fundus photography and ocular examinations from June 2002 to November 2002 were included in this retrospective study. The patients, recruited from a hospital-based, retina referral practice, underwent 45 degrees, non-mydriatic, digital fundus photography using a non-mydriatic fundus camera. One image was obtained focusing the mid fundus between the optic disc and the macula. The fundus images were printed and graded by endocrinologists and a retinal specialist separately. The patients also underwent complete standard ocular examinations as the reference method for determining diabetic retinopathy, including dilation of their pupils and slit-lamp biomicroscopy done by ophthalmologists. The sensitivity and specificity of the digital photographic method were calculated by comparison to the reference method.

RESULTS

The sensitivity and specificity of the retinal specialist's diabetic retinopathy grades were 53.8 and 89.0%, respectively. The sensitivity and specificity of the endocrinologists' grades were 45 and 75.3%, respectively. The false negative rates were 22 and 21.5% for endocrinologists and the retinal specialist, respectively.

CONCLUSIONS

Screening for diabetic retinopathy using one-field, non-mydriatic, 45 degrees digital photography is inadequate.

The effect of digital image resolution and compression on anterior eye imaging

AIM

To determine the theoretical and clinical minimum image pixel resolution and maximum compression appropriate for anterior eye image storage.

METHODS

Clinical images of the bulbar conjunctiva, palpebral conjunctiva, and corneal staining were taken at the maximum resolution of Nikon:CoolPix990 (2048x1360 pixels), DVC:1312C (1280x811), and JAI:CV-S3200 (767x569) single chip cameras and the JVC:KYF58 (767x569) three chip camera. The images were stored in TIFF format and further copies created with reduced resolution or compressed. The images were then ranked for clarity on a 15 inch monitor (resolution 1280 x 1024) by 20 optometrists and analysed by objective image analysis grading. Theoretical calculation of the resolution necessary to detect the smallest objects of clinical interest was also conducted.

RESULTS

Theoretical calculation suggested that the minimum resolution should be > or = 579 horizontal pixels at 25x magnification. Image quality was perceived subjectively as being reduced when the pixel resolution was lower than 767 x 569 (p<0.005) or the image was compressed as a BMP or <50% quality JPEG (p<0.005). Objective image analysis techniques were less susceptible to changes in image quality, particularly when using colour extraction techniques.

CONCLUSION

It is appropriate to store anterior eye images at between 1280 x 811 and 767 x 569 pixel resolution and at up to 1:70 JPEG compression.

Effect of video display on the grading of diabetic retinopathy

PURPOSE

Digital images are increasingly being used in ophthalmology. These may be viewed either on thin-film transistor (TFT) or on cathode-ray tube (CRT) displays. However, there is little data showing which is superior. In this study, we compared the performance of CRT and TFT displays for grading of both compressed and uncompressed images of diabetic retinopathy.

METHODS

A total of 49 35 mm transparencies of diabetic retinopathy were scanned and compressed. The images comprised 17 with no retinopathy, eight with background, five with preproliferative, and 19 with proliferative retinopathy. Four levels of compression were used: 0, 70, 80, and 90%. A total of 196 randomised images were presented to two masked graders using both TFT and CRT displays under uniform lighting conditions, 2 months apart. The grade of retinopathy was assessed. Statistical analysis of grading accuracy was performed using receiver operator characteristic curves of sensitivity and specificity and the Stuart-Maxwell test for paired, nonparametric data.

RESULTS

Both displays showed high sensitivity and specificity for the detection of any retinopathy. For the specific grade of retinopathy, the CRT performed slightly better with a sensitivity of 0.80 for uncompressed images, compared with 0.75 using the TFT. Compression reduced these sensitivities to 0.73 on the CRT and 0.63 on the TFT. Grading of uncompressed images magnified to four times their original size was more accurate on the TFT.

CONCLUSIONS

Grading on both displays met sensitivity and specificity criteria proposed by Diabetes UK (formerly British Diabetic Association) for screening of diabetic retinopathy. The CRT generally performed slightly better than the TFT in relation to the detection of the specific grade of retinopathy.

A national retinal screening programme for diabetes in Scotland

The Health Technology Board Scotland (HTBS) have issued recommendations for eye screening in patients with diabetes. These are based on evidence-based clinical studies. Evidence-based studies do not answer all the practical issues, and some conclusions have thus been extrapolated from the known evidence base. Other factors such as patient issues, organizational issues and cost effectiveness have also been incorporated into the recommendations. HTBS recommend single-field digital retinal photography. Retinal photography best addresses the issues of adequate sensitivity and accountable quality assurance. Non-mydriatic photography is recommended, followed by immediate use of dilating eye drops if it is unsuccessful, followed by slit-lamp examination if both of these approaches fail. An independent grading scheme has been established, which is similar to the 'Global', and compatible with the National Screening Committee (NSC) grading scheme. The rationale for these recommendations, and debate behind some of the decisions, is laid out in this article.

Is digital image compression acceptable within diabetic retinopathy screening?

AIMS

The National Screening Committee (NSC), whilst recommending the use of digital mydriatic retinal photography for diabetic retinopathy screening, has not yet accepted the use of digitally compressed images for grading. By greatly reducing the file size, however, compression of images is invaluable for storage and for its rapid transmission across computer networks. We undertook a study to compare the different levels of JPEG compression with the original bit-mapped image to determine whether there was any loss of clinical detail following compression.

METHODS

Three hundred and thirty images were analysed in this study. These images had been captured from 66 eyes consecutively photographed in a diabetic retinopathy screening programme, using a Sony DXC-950 P 3CCD colour video camera mounted on a Canon CR6-45NMf fundus camera. Single 45 degrees macula-centred images were taken from each eye. The images were compressed using the JPEG algorithm within Adobe Photoshop (version 4.0) and then displayed with a Sony Trinitron colour monitor. Four different levels of compression were used, JPEG-1, JPEG-2, JPEG-3, JPEG-4, and an objective analysis was undertaken using 'lesion counts'. The compressed images were assessed separately and blindly and the results compared with their original BMP images.

RESULTS

Eight BMP images could not be evaluated (five right eye and three left eye). A total of 290 images were therefore used in the final evaluation. All the JPEG-1 images with file sizes between 16 and 24 kb were found to be 'pixelated', while the JPEG-4 images (66-107 kb) appeared similar to the original BMP (1.3 Mb) images. Both JPEG-2 and JPEG-3 images had significantly lower counted lesions than the BMP images.

CONCLUSIONS

From our findings we can conclude that only some degree of image compression (compression ratios of 1 : 20 to 1 : 12) with file sizes of 66-107 kb is permissible using JPEG format, whereas the images obtained after higher compression ratios may not be suitable for diabetic retinopathy screening.

Endoscopic data acquisition and storage

Recent developments in the field of digital image acquisition and storage provide new opportunities for recording endoscopic procedures for use in the medical record and remote access for education and consultation. This article reviews the recent literature in the field of endoscopy.