External quality assurance for image grading in the Scottish Diabetic Retinopathy Screening Programme.

AIMS: To develop and evaluate an image grading external quality assurance system for the Scottish Diabetic Retinopathy Screening Programme. METHOD: A web-based image grading system was developed which closely matches the current Scottish national screening software. Two rounds of external quality assurance were run in autumn 2008 and spring 2010, each time using the same 100 images. Graders were compared with a consensus standard derived from the top-level graders' results. After the first round, the centre lead clinicians and top-level graders reviewed the results and drew up guidance notes for the second round. RESULTS: Grader sensitivities ranged from 60.0 to 100% (median 92.5%) in 2008, and from 62.5 to 100% (median 92.5%) in 2010. Specificities ranged from 34.0 to 98.0% (median 86%) in 2008, and 54.0 to 100% (median 88%) in 2010. There was no difference in sensitivity between grader levels, but first-level graders had a significantly lower specificity than level-two and level-three graders. In 2008, one centre had a lower sensitivity but higher specificity than the majority of centres. Following the feedback from the first round, overall agreement improved in 2010 and there were no longer any significant differences between centres. CONCLUSIONS: A useful educational tool has been developed for image grading external quality assurance.

Costs and consequences of automated algorithms versus manual grading for the detection of referable diabetic retinopathy.

AIMS: To assess the cost-effectiveness of an improved automated grading algorithm for diabetic retinopathy against a previously described algorithm, and in comparison with manual grading. METHODS: Efficacy of the alternative algorithms was assessed using a reference graded set of images from three screening centres in Scotland (1253 cases with observable/referable retinopathy and 6333 individuals with mild or no retinopathy). Screening outcomes and grading and diagnosis costs were modelled for a cohort of 180 000 people, with prevalence of referable retinopathy at 4%. Algorithm (b), which combines image quality assessment with detection algorithms for microaneurysms (MA), blot haemorrhages and exudates, was compared with a simpler algorithm (a) (using image quality assessment and MA/dot haemorrhage (DH) detection), and the current practice of manual grading. RESULTS: Compared with algorithm (a), algorithm (b) would identify an additional 113 cases of referable retinopathy for an incremental cost of pound 68 per additional case. Compared with manual grading, automated grading would be expected to identify between 54 and 123 fewer referable cases, for a grading cost saving between pound 3834 and pound 1727 per case missed. Extrapolation modelling over a 20-year time horizon suggests manual grading would cost between pound 25,676 and pound 267,115 per additional quality adjusted life year gained. CONCLUSIONS: Algorithm (b) is more cost-effective than the algorithm based on quality assessment and MA/DH detection. With respect to the value of introducing automated detection systems into screening programmes, automated grading operates within the recommended national standards in Scotland and is likely to be considered a cost-effective alternative to manual disease/no disease grading.

Cost-effectiveness of implementing automated grading within the national screening programme for diabetic retinopathy in Scotland.

AIMS: National screening programmes for diabetic retinopathy using digital photography and multi-level manual grading systems are currently being implemented in the UK. Here, we assess the cost-effectiveness of replacing first level manual grading in the National Screening Programme in Scotland with an automated system developed to assess image quality and detect the presence of any retinopathy. METHODS: A decision tree model was developed and populated using sensitivity/specificity and cost data based on a study of 6722 patients in the Grampian region. Costs to the NHS, and the number of appropriate screening outcomes and true referable cases detected in 1 year were assessed. RESULTS: For the diabetic population of Scotland (approximately 160,000), with prevalence of referable retinopathy at 4% (6400 true cases), the automated strategy would be expected to identify 5560 cases (86.9%) and the manual strategy 5610 cases (87.7%). However, the automated system led to savings in grading and quality assurance costs to the NHS of 201,600 pounds per year. The additional cost per additional referable case detected (manual vs automated) totalled 4088 pounds and the additional cost per additional appropriate screening outcome (manual vs automated) was 1990 pounds. CONCLUSIONS: Given that automated grading is less costly and of similar effectiveness, it is likely to be considered a cost-effective alternative to manual grading.

The efficacy of automated "disease/no disease" grading for diabetic retinopathy in a systematic screening programme.

AIM: To assess the efficacy of automated "disease/no disease" grading for diabetic retinopathy within a systematic screening programme. METHODS: Anonymised images were obtained from consecutive patients attending a regional primary care based diabetic retinopathy screening programme. A training set of 1067 images was used to develop automated grading algorithms. The final software was tested using a separate set of 14 406 images from 6722 patients. The sensitivity and specificity of manual and automated systems operating as "disease/no disease" graders (detecting poor quality images and any diabetic retinopathy) were determined relative to a clinical reference standard. RESULTS: The reference standard classified 8.2% of the patients as having ungradeable images (technical failures) and 62.5% as having no retinopathy. Detection of technical failures or any retinopathy was achieved by manual grading with 86.5% sensitivity (95% confidence interval 85.1 to 87.8) and 95.3% specificity (94.6 to 95.9) and by automated grading with 90.5% sensitivity (89.3 to 91.6) and 67.4% specificity (66.0 to 68.8). Manual and automated grading detected 99.1% and 97.9%, respectively, of patients with referable or observable retinopathy/maculopathy. Manual and automated grading detected 95.7% and 99.8%, respectively, of technical failures. CONCLUSION: Automated "disease/no disease" grading of diabetic retinopathy could safely reduce the burden of grading in diabetic retinopathy screening programmes.

A reference standard for the measurement of macular oedema.

Macular oedema is associated with several conditions that lead to blindness. Accurate measurement of macular thickness is important in order to follow disease progression and evaluate treatments. Four techniques are examined to determine the best reference standard for the detection and quantification of macular oedema: ultrasound, optical coherence tomography, the retinal thickness analyser, and the scanning laser ophthalmoscope. The three optical techniques have the highest resolution and sensitivity, in particular optical coherence tomography. Ultrasound can be useful where dense opacities preclude optical imaging.

A comparative evaluation of digital imaging, retinal photography and optometrist examination in screening for diabetic retinopathy.

AIMS: To compare the respective performances of digital retinal imaging, fundus photography and slit-lamp biomicroscopy performed by trained optometrists, in screening for diabetic retinopathy. To assess the potential contribution of automated digital image analysis to a screening programme. METHODS: A group of 586 patients recruited from a diabetic clinic underwent three or four mydriatic screening methods for retinal examination. The respective performances of digital imaging (n=586; graded manually), colour slides (n=586; graded manually), and slit-lamp examination by specially trained optometrists (n=485), were evaluated against a reference standard of slit-lamp biomicroscopy by ophthalmologists with a special interest in medical retina. The performance of automated grading of the digital images by computer was also assessed. RESULTS: Slit-lamp examination by optometrists for referable diabetic retinopathy achieved a sensitivity of 73% (52-88) and a specificity of 90% (87-93). Using two-field imaging, manual grading of red-free digital images achieved a sensitivity of 93% (82-98) and a specificity of 87% (84-90), and for colour slides, a sensitivity of 96% (87-100) and a specificity of 89% (86-91). Almost identical results were achieved for both methods with single macular field imaging. Digital imaging had a lower technical failure rate (4.4% of patients) than colour slide photography (11.9%). Applying an automated grading protocol to the digital images detected any retinopathy, with a sensitivity of 83% (77-89) and a specificity of 71% (66-75) and diabetic macular oedema with a sensitivity of 76% (53-92) and a specificity of 85% (82-88). CONCLUSIONS: Both manual grading methods produced similar results whether using a one- or two-field protocol. Technical failures rates, and hence need for recall, were lower with digital imaging. One-field grading of fundus photographs appeared to be as effective as two-field. The optometrists achieved the lowest sensitivities but reported no technical failures. Automated grading of retinal images can improve efficiency of resource utilization in diabetic retinopathy screening.

Measurement of cerebral perfusion volume and 99mTc-HMPAO uptake using SPECT in controls and patients with Alzheimer's disease.

Methods for quantifying the changes in brain function observed in single photon emission computed tomography (SPECT) using hexamethylenepropylene amine oxime (HMPAO) for patients with Alzheimer's disease have the potential of improving the diagnostic accuracy of the procedure and its ability to monitor response to treatment. The absolute percentage uptake of HMPAO and the cerebral perfusion volume (CPV) of the brain were assessed using SPECT in 26 patients with mild to moderate Alzheimer's disease (AD) and 24 control subjects. A subset of 15 control subjects, which was age-matched to the AD patients, was selected to allow fair statistical comparison of parameters between groups. The percentage of brain volume with reduced perfusion (R) and a volume loss index (VLI), given by /CPV, were also calculated. Eight of the control subjects were studied on a second occasion after a mean period of 6 months. There was no significant difference in percentage uptake between controls and AD patients, the mean value being 5.8%. Cerebral perfusion volume in controls was found to depend on sex (mean value in males and females being 1327 ml and 1222 ml, respectively) and on age. The volume loss index corrected for age and sex provided good discrimination between controls and AD subjects giving a sensitivity and specificity of 81% and 96%, respectively. The repeatability coefficient, the 95% confidence limit for the difference between repeat measurements, on controls was 67 ml (5%). The measurement of cerebral perfusion volume and related indices may be of value in identifying patients with early Alzheimer's disease and in following their response to treatment.

A comparison of performance of three gamma camera systems for positron emission tomography.

The development of coincidence positron imaging using a gamma camera should significantly increase PET utilization. This study has compared the performance of three such systems with each other and with an equivalent single photon emission computed tomography (SPECT) measurement. A Carlson phantom, which consisted of a 20 cm diameter cylinder containing areas of uniform activity, hot lesions, cold lesions and orthogonal alternate hot and cold lines, was filled with an activity of 18F chosen to try and best exploit each camera and imaged for a fixed time (27 min). The image quality of 9 mm thick slices in each section of the phantom was compared visually. Several image quality parameters were also compared including line source resolution and noise equivalent count rate. There were considerable differences in PET image quality between the three cameras but all were at least as good as the SPECT image. The full width at half maximum (FWHM) resolution of all systems was similar (approximately 4.5 mm) but the maximum noise equivalent count rates in a 20 cm cylinder were significantly different (6.3, 2.6 and 1.6 kcps) (where cps is counts per second) and correlated with the phantom image quality.

Semi-automated tabulation of the 3D topology and morphology of branching networks using CT: application to the airway tree.

Detailed information on biological branching networks (optical nerves, airways or blood vessels) is often required to improve the analysis of 3D medical imaging data. A semi-automated algorithm has been developed to obtain the full 3D topology and dimensions (direction cosine, length, diameter, branching and gravity angles) of branching networks using their CT images. It has been tested using CT images of a simple Perspex branching network and applied to the CT images of a human cast of the airway tree. The morphology and topology of the computer derived network were compared with the manually measured dimensions. Good agreement was found. The airways dimensions also compared well with previous values quoted in literature. This algorithm can provide complete data set analysis much more quickly than manual measurements. Its use is limited by the CT resolution which means that very small branches are not visible. New data are presented on the branching angles of the airway tree.