Home page of
Frédéric Zana

Centre de Morphologie Mathématique
Ecole Nationale Supérieure des Mines de Paris
35, rue Saint Honoré
77305 Fontainebleau,
France


CLICK HERE FOR A DEMO OF RETINAL IMAGE REGISTRATION 


During my four-year study at The Ecole Normale Superieure, I have graduated in pure mathematics (1993) and I have completed a Master of Science in Computer Science (1994) and a Master of Science in Macroeconomics (1996). I have also spent height months in Utah (U of SLC, as part of my MS in Computer Science), where I have modeled manufacturing tolerance (for ARPA Worshop of image understanding 1994) using an original Spline curve: Interval Splines (results are presented in the IEEE trans. on Robo. & Auto. paper).

I have participated to a group of discussion concerning a social analysis of the Internet (atelier Internet). My contribution Internet or interactive television (in French) was made a (small) part of a report to the French Ministry of Education.

I have began a Ph.D. in October 1996 at the Center of Mathematical Morphology of the Paris School of Mines. It has been completed in july 1999, with highest honors and congratulations.


Research Subject

Diabetes is the first cause of legal blindness in Westerns countries. It causes different kinds of diabetic retinopathies, related with vascular abnormalities of different gravity. The earliest retinal abnormalities caused by diabetes are microaneurysms (vascular lesions that appear as dots), the growth of the foveal avascular zone (which is the zone of central vision), and leakage. The lesions causing the abnormalities are very small (down to 10µm), therefore it is necessary to inject a dye in order to enhance the contrast of vessels. The specific dye commonly used to enhance retinal vessel is the Fluorescein. The evolution of the abnormalities is related to the evolution of the illness. Tracking small dots is a time consuming task for physicians. Digitized fluorescein angiography will allow automatic counting of microaneurysms (or other abnormalities), and image registration is necessary to compute the evolution of parameters like the appearing/disappearing of microaneurysms, the size of the foveal avascular zone, and the evolution of other abnormalities.

The goal of our research team is to develop algorithms that detect some abnormalities on digitized angiographies, eye fundus or green images, and then register them to help physician follow closely the evolution of the illness. Our focus is not limited to the case of diabetes.

My main research project consists in analysis and registration of angiographic images of the retina. I plan to measure some useful parameters, then follow their evolution between medical examinations in order to facilitate diagnosis of various diseases. Up to now, I have worked on microaneurysm and avascular zone detection, which are useful parameters in case of early diabetic retinophathy and macular degenerescence. I have also detected necrosis caused by cytomegalovirus that appear as a mighty consequence of AIDS. This study will lead to an internet server for retinal image analysis. The tasks required for my Ph.D. were the following:

  • Vessels and medical parameters extraction
  • extraction of bifurcation points from the image of vessels
  • perform non rigid registration of images
  • compute dynamical parameters
  • example of treatment

    RESULTS:

    I have performed a robust algorithm for vessel segmentation and microaneurysm detection (presented in DSP97, and in the Colloque Mines Fontainebleau, and to be published in IEEE Trans. on Image Processing). I have also realized an automatic matching on a database of pairs of images, based on the segmentation of vessels (published in IEEE Trans. on Medical Imaging). This algorithm uses the former vessel segmentation to detect bifurcation points that are used to perform registration. This strategy allows multimodal registration, temporal registration, or both. In case of diabetic retinopathy, it was a real challenge since the vascular tree is altered. Our algorithm thus deals with occlusion and/or misdetection.

    Other parameters have been extracted, like the avascular zone in angiographic images (article presented in ISMM 98) , neovascularisation visible under ICG, and the capillary net around the macula for Florida cases. A study concerning cytomegalovirus detection was done for de la Tymone Hospital.


    More detailed results:

    I. Robust Segmentation of Vessels from Retinal Angiography

    (French written short presentation aimed at non-specialist)

    Postscript version of DSP97 paper (English written)

    II. Registration of Retinal Images

    (detailed presentation, English written, with several examples)

    multimodal registration: angiography and green image

    temporal registration: between angiographies

    temporal registration: between green images

    temporal and multimodal registration: angiography and green image

    registration with few information: angiography and green image of poor quality

    registration with occlusion: green image and angiography with missing parts

    III. Macula Segmentation using Region Merging

    (English written abstract of ISMM'98 paper)

    Postscript version of ISMM98 paper


    Bibliography

    @InProceedings {Za97, Author = { F. Zana and J.C. Klein}, Title = { Robust Segmentation of Vessels from Retinal Angiography}, Booktitle = { International Conference on Digital Signal Processing }, Month = {July}, Address = {Santorini, Greece}, Year = {1997}, Pages = {1087-1091} }

    @InProceedings {Za98, Author = { F. ZANA , I. MEUNIER , and J.C. KLEIN }, Title = { A REGION MERGING ALGORITHM USING MATHEMATICAL MORPHOLOGY: APPLICATION TO MACULA DETECTION }, Booktitle = { Mathematical Morphology and its Applications to Image and Signal Processing }, Month = {June}, Address = {Amsterdam, Holland}, Year = {1998}, Pages = {423-430} }

    @InProceedings {Za99, Author = { F. ZANA , and J.C. KLEIN }, Title = { A Registration Algorithm of Eye Fundus Images using a Bayesien Hough Transform.}, Booktitle = { Image Processing and its Applications}, Month = {July}, Address = {Manchester, UK}, Year = {1999}, Pages = {} }

    @ARTICLE{Za99, Author = { F. Zana and J.C. Klein }, Title = { Segmentation of Vessel-Like Patterns using Mathematical Morphology and Curvature Evaluation}, Journal = {proposed to IEEE Transaction on Image Processing }, volume = {}, Number = {}, Year = {}, Pages = {} }

    @ARTICLE{Za99, Author = { F. Zana and J.C. Klein }, Title = { A Multi-Modal Registration Algorithm of Eye Fundus Images Using Vessels Detection and Hough Transform }, Journal = {IEEE Transaction on Medical Imaging }, volume = {18}, Number = {5}, Year = {1999}, Pages = {} }


      Frédéric Zana
      born the 9th of May 1972
      35 rue Saint-Honoré
      77 305 Fontainebleau, FRANCE