“Automated retinal image analysis (ARIA) for diabetic retinopathy in telemedicine”: are we ready?

This week’s task for #MI227, Clinical and Laboratory Information Systems for #MSHI required us to find an article describing the adoption or use of an EMR system, a CPOE system, a medication administration system, a telemedicine system, a telehealth system, a PHR, or other clinical or laboratory information system or application, discuss its key points, lessons learned and how this can relate to the Philippines. I chose

Automated retinal image analysis for diabetic retinopathy in telemedicine by Sim et al, published online in Curr Diab Rep (2015) 15:14. https://www.researchgate.net/publication/272518326_Automated_Retinal_Image_Analysis_for_Diabetic_Retinopathy_in_Telemedicine

Key points:

The article by Sim et al discusses how an automation process analysis can negate the heavy reliance of current telemedicine practices on specially trained retinal image graders, thereby improving the delivery of diabetes eye care, expediting diagnosis and facilitating referral to a treatment facility. In addition, the potential of such a system integrating with the electronic medical record permits a more accurate analysis and prognostication of the disease.

The need for early diagnosis (and necessary intervention) is diabetic retinopathy is recognized as a major factor in reducing blindness due to this pathology. In England and Wales where systematic population based screening is in place, diabetic retinopathy is no longer the leading cause of blindness. Yet, even in the developed countries, such as the US for example, access to eye care is only 60-90%, and is presumably much lower elsewhere. The ARIA potentially can distribute quality eye care or screening to virtually anywhere, with the software providing automated image analysis algorithms.

It is the need for “trained personnel for image reading and grading a large volume of retinal images” that the ARIA addresses, making the reading process less dependent on humans. Human readers, however, will continually be required for quality control, arbitration, and interpretation of atypical retinal images. The ARIA is envisioned to be linked to a patient’s electronic medical record.

ARIA was developed to perform computer algorithms capable of computer-aided detection (CADe) and computer-aided diagnosis (CADx). ARIA addressed two issues: image quality assessment and image analysis. Image quality assessment required that pre-processing improve on factors affecting image quality such as brightness, contrast, signal/noise ratio, and/or determining image clarity by assessing vessels around the macula. Image analysis begins with initial segmentation or the identification and localization of normal anatomy so that the “normal” is excluded from image analysis of what is pathologic (microaneurysms, exudates, hemorrhage, beading, neovascularization). The challenge encountered with such a system was how ARIA could deal with distractors such as retinal capillaries, choroidal vessels, and reflection artifacts.

ARIA systems currently deployed in telemedicine and screening programs include the iGradingM, The Triad Network, Iowa Detection Program (IDx-DR), RetmarkerDR, and Retinalyze System.

Future development of ARIA algorithms require a set of images used for calibration and “training” where human labeled sample images are used to teach the computer to remember such an image and its reading. Two public datasets are available for such use, the Methods for Evaluating Segmentation and Indexing techniques Dedicated to Retinal Ophthalmology (MESSIDOR) and Retinopathy Online Challege (ROC). More could be made available if only regulatory and proprietary barriers could be breached.

Lessons Learned

Telemedicine programs done right, can fill a void by exponentially increasing the capabilities of performing early screening and detection (with computer-aided detection and diagnosis in the case of ARIA), virtually eliminating boundaries. To quote the article by Sim et al “telemedicine programs for diabetic retinopathy should include:

1. Remote, reliable, cost effective image acquisition system
2. An image reading center
3. A clinical recording center that:
4. communicates results to physicians and patients
5. facilitates appointments for follow-up assessments
6. facilitates appointments for treatment
7. IT and technical support
8. Administrative support
9. Trained personnel for image reading and grading a large volume of retinal images”
10. Telemedicine concerns should address ethical and patient privacy issues.

Telemedicine diagnosis is still continually evolving and changing and requires research, validation, revalidation, improvement in software, image capture, and image sharing. Human intervention still cannot be fully removed as human readers will continually be required for quality control, arbitration, and interpretation of atypical retinal images.

Finally, collaboration is required at different levels:

• National health authorities, commercial companies, and the medical profession
• Large repositories of real-life datasets for expert-annotated images
• International agreement on performance criteria and evaluation and validationparameters
• Involvement of primary care providers

What this means for the Philippines

In the Philippines, there remains a heavy reliance of screening on trained specialists in Retina, a human resource that is concentrated in the urban areas such as the National Capital Region and wanting in the provinces and outskirts of the metropolis and most especially in the remote areas. Using telemedicine partially solves the problem as reading centers are staffed by these trained Retina specialists. Specialists staff reading centers and commit to evaluating photographs within a predetermined time. This minimizes the cost of unneccessary travel to urban centers, and includes the primary care physician in diabetes care. We are at this stage.

Automating the process of reading can exponentially increase the number that can be screened, in real time. Readings can be provided to the health care team, almost immediately after a patient performs the test. This frees up the the retina specialist to deal more with patients for counseling, education, performings lasers, or surgeries.

We have problems to deal with, the infamous slow internet connection precludes real time image transfer. Patient privacy issues remain a concern. Multilevel collaboration, interhospital collaboration, data sharing are still wanting. Diabetic Retinopathy ranks among the leading cause of vision impairment in the Pacific. (ref 2) The World Health Organization estimates that 15% of blindness is due to diabetic retinopathy or glaucoma in the Western Pacific region. If telemedicine can take off, if screening can exponentially increase, and intervention introduced early, this number can drastically reduce a significant cause of blindness.

References:

1. Sim DA, Keane PA, Tufail A, et al. Automated Retinal Image Analysis for Diabetic Retinopathy in Telemedicine. Curr Diab Rep (2015) 15: 14 https://www.researchgate.net/publication/272518326_Automated_Retinal_Image_Analysis_for_Diabetic_Retinopathy_in_Telemedicine Accessed January 28, 2016.

2. Keefe JE, Konyama K, Taylor HR. Visiion Impairment in the Pacific Region. Br J Ophthalmology 2002: 86(6). 605-610. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1771168/ Accessed January 28, 2016.

3. World Health Organization. Vision 2020 report. http://www.who.int/blindness/Vision2020_report.pdf Accessed January 28, 2016.