Multi-Modality Medical Image Registration for Eye Diagnosis
Project Type: General Research Project
Research Field: Multisensor Fusion and Multirobot Systems for 3D Reconstruction
Sponsors: POSI/SRI/45151/2002
Time span: 02/2004-06/2006

Diabetic retinopathy is the leading cause of legal blindness in developed countries. Ten percent of the EU population are expected to develop diabetes by the year 2010 while in the USA there are, currently 100.000 new cases each year. With the incidence of diabetic retinopathy among the diabetic population, a large socio-economic impact is expected.

Each method of diagnosis focuses on a particular feature of the retina. This gives a partial view of the changes occurring in the retina in a given time whereas an entire overview is essential for a better diagnosis and to understand the changes that are occurring and are expected to occur in the near future.

In order to achieve a broader overview of the changes occurring in the retina, it is necessary to correlate several complementary examinations: fluorescein angiography, confocal scanning laser angiography, retinal blood flow, retinal thickness and retinal leakage.

Registration of the modalities involved has to be automatic so that it can be used in the daily practice in the clinical environment. Differences in scale, resolution, distortions, rotation and translations must be taken in account.

Registration methods are classified as feature or intensity based.

The former is based on the segmentation of homologous geometric landmarks present in the images while intensity based registration searches a space transformation for the maximum of a given measure based on a particular criteria of evaluation of similarity. This similarity is classified as: identity relationship; affine relationship; functional relationship; and statistical relationship.

A physical model will be established so that the knowledge on structures being measured can be incorporated in the process. Whereas the image corresponds to light coming from a layer of given depth (different at each point), for retinography the information gathered from the same retina with retinal flowmeter takes into account a light from different layers due to the confocality of the system. This is to say that different entities are being measured with different methodologies.

A physical model will be established to provide information on what to expect from each modality, its relation with the whole reality, and what to extent the information is related and in which aspects.

The approach to be taken in this project is to test several registration techniques for each modality. The techniques to be tested are the ones that most likely shown any relation to the physical model previously developed.

In order to address all the above-mentioned problem the physical model will be established incorporating the knowledge of the physical eye and the acquisition technology, as well as its statistical properties.

Related People

Jorge Dias
João Filipe Ferreira
Carlos Miguel Mendes da Silva
Rui Bernardes