Spectral imaging could detect diabetic retinopathy earlier

Text Ulla Kaltiala I Photos Varpu Heiskanen

Diabetes is a growing global health concern, but millions of people have diabetes without knowing it. Diabetic retinopathy, a common complication of the disease and a leading cause of blindness in adults, can also progress for years without any warning signs. New technology based on spectral imaging could help in the early detection of both.

Researchers in the UEF Colour Research group have developed a prototype of a new spectral eye fundus camera which can detect retinopathic abnormalities more accurately and at an earlier stage than the present clinical imaging methods. At the same time, they have created a unique spectral image database of diabetic eye fundi.

“The amount of information in traditional RGB colour images of the retina is very limited compared to our method. An RGB image consists of only three color channels – red, green and blue – whereas with our spectral fundus camera, we can take images with 30 spectral colour channels,” says MSc Pauli Fält, a postgraduate student in physics who has developed the method together with laboratory engineer, PhD Jouni Hiltunen.

“Different objects of the retina can be emphasized with different wavelengths of light, perceived as different colors. Based on this we can choose the optimal colour channels to make retinal changes much more visible in the image,” Mr Fält explains.

As an example, he shows two images of a fundus with fibrotic changes. In the RGB image the changes are hardly visible to the untrained eye, but the one with optimal spectral colour channels shows them with impressive contrast and clarity. Which of these images would you like to use as the basis of your diagnosis?

Spectral images can’t be taken at the touch of a button, though. The patient’s eye fundus has to be illuminated and photographed with each of the 30 colours one by one, which takes minutes.

On the left: In this standard RGB-image of a retina, the abnormal fibrotic growth appears a fuzzy mist. On the right: The visibility of the fibrosis can be enhanced by using spectral color information.

On the left: RGB-image of a retina with diabetic retinopathy. On the right: Visibility and contrast can be improved by using optimal spectral color channels.

Tools for diagnostics and follow-up

These new insights into fundus imaging are the result of a multidisciplinary project called ImageRet – Optimal Detection and Decision-support Diagnosis of Diabetic Retinopathy. The three-year project which was completed at the end of 2009 was mainly funded by the Finnish National Agency for Technology and Innovation. The goal of the project was to develop new imaging hardware and novel image processing methods to assist in medical decision making in diabetic retinopathy diagnosis.

The partners in the project included the University of Eastern Finland, Lappeenranta University of Technology (LUT), Mikkeli Polytechnic, University of Bristol, and several companies.

The idea of automated diagnosis of diabetic retinopathy has been intriguing researchers of medical image processing, such as the LUT Machine Vision and Pattern Recognition Laboratory led by Professor Heikki Kälviäinen. Automated methods would make more extensive screenings possible. Retinal image databases have already been created which researchers can make use of in developing pattern recognition methods and algorithms for the detection of retinal structures and abnormalities.

The spectral image database of diabetic eye fundi, however, is the first of its kind in the world. It was created as a joint effort between the Colour Research group and the UEF Unit of Ophthalmology. Spectral images of the retinas of 54 voluntary diabetic patients and 12 control subjects were collected. The images demonstrate different signs and severities of diabetic retinopathy, accompanied by expert annotations.

“The database will be made public for researchers in the near future. It will also be augmented with additional data and annotations. In potential automated methods using the database, patients’ fundus images could be compared to a statistical model based on the expert annotations,” Dr Hiltunen says.

The spectral fundus camera could actually be useful in the diagnosis and follow-up of a variety of eye diseases; macular degeneration and glaucoma are some examples. The new technology is already being transferred to commercial product development.

“More generally, there are many potential medical applications for spectral imaging, such as the imaging of skin diseases,” Mr Fält adds. “Along with the development of spectral cameras and the growing evidence of the advantages of spectral data, this type of imaging will probably gain popularity in medical use.”

Spectral colour research – a spectrum of possibilities

The University of Eastern Finland is one of the world leaders in spectral colour research. The Colour Research group, headed by Professors Timo Jääskeläinen and Jussi Parkkinen, has been focusing on spectral colour research since the 1980s, combining the university’s expertise of physics and computer science in a unique manner. In recent years, industrial cooperation has expanded. Since 2003, the group has been part of the InFotonics Center Joensuu, a service centre designed to facilitate the commercialization of high technology research.

But what is this spectrum thing all about? “The most accurate way to represent the colour of an object is to measure the spectrum of the electromagnetic radiation sent by the object,” says Docent Markku Hauta-Kasari, Director of the InFotonics Center.

The electromagnetic radiation he’s talking about is better known to us mere mortals as colours: we perceive different wavelengths of visible light as different colours. However, the research group applies spectral methods to electromagnetic radiation beyond visible light as well, like ultraviolet light and infrared radiation. Their Colour Research Laboratory is equipped with up-to-date devices for accurate colour measurements and analysis. Once collected, spectral data is processed with computational methods.

“The advantage of spectral imaging is the amount of additive information it captures, which allows for an accurate identification of chemical and optical qualities and structures of the object, among other things.”

Applications for spectral colour research are numerous and cover a variety of fields, from forestry, mining and metal industry to paper and plastics industries, to name a few. In domestic and international industrial cooperation projects, spectral imaging systems are often used for quality control and improvement. In addition to long-term R&D projects, the group carries out ad hoc feasibility studies and concept development for companies.

“Our research will focus more and more on medical applications. In an ongoing project we aim to create a new type of brain surgery workstation by combining a spectral video camera with a surgery microscope. In another one, we are developing a spectral imaging method for the diagnosis of cartilage degeneration.”

There are interesting prospects in food industry as well.“ We are currently starting work related to a method for detecting the contamination of food products. Using UV light, we are interested in studying whether different bacteria can be detected without time-consuming culture techniques by means of different fluorescence properties.”

There is active international cooperation and researcher exchange with other research groups in the field and especially with Japanese universities. On June 14−17, 2010 the group will host the Fifth European Conference on Colour in Graphics, Imaging, and Vision in Joensuu.