Abstract
Purpose: :
Large scale screening for diabetic retinopathy was implemented commercially in diabetic patients who were visiting a primary care office and had not been seen by an eye care specialist in the previous year. As expected, some patients needed referral due to unreadable images. However, in many cases the readers judged that poor image quality could have been avoided and should be re-imaged. Such cases were reviewed in an attempt to determine the causes and suggest means to reduce their occurrence.
Methods: :
Between July 2007 and June 2008 15,678 diabetic patients were screened with the DigiScope (EyeTel Imaging, Inc). Staff members were trained by a multi-media presentation embedded in the instrument and/or by a co-worker who had received a short instruction when the instrument was first placed. Imaging was performed under mydriasis as recommended for such an exam. Cases marked by readers as ‘re-image’ were retrieved from the database and a decision tree was used to identify the main cause for poor imaging.
Results: :
Fundus images were obtained at 356 sites by 856 office staff members. Forty three percent of patients screened were 60 or more years old. A random sample of 97 cases out of 2070 (12% of total) marked re-image was studied. In 13 cases the cause of poor imaging could not be determined and in another 6 it was attributed to the failure of two instruments prior to their replacement. In 91% of the 78 remaining cases, the first eye imaged triggered the classification. The causes were identified as operator error (53%) , lack of patient fixation on target (42%), pupil size (3%), eyelid obstruction (1%) and miscellaneous factors (1%). Out of the 53% operator related failures, 30% occurred while setting the fundus focus and 23% while setting the pupil working distance.
Conclusions: :
The results indicate that the 12% occurrence of preventable poor imaging potentially could be reduced by an automated algorithm that provides, during the imaging session, quality assurance feedback to the operator and the patient. These results may not apply to non-mydriatic imaging.
Keywords: image processing • diabetic retinopathy • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)