Research in Multimodal Imaging in Ophthalmology

  • Fully automatic segmentation of the Eye in 3D MRI [Published]:
Picture
a) Automatic detection of the eye. b) Cropping of the Region of interest and alignment of the MRI Volume. c) Automatic segmentation of the Eye for the lens, the sclera and cornea and the vitreous humor.
Delineation of ocular anatomy is a tedious task, however it is required for treatment planning and for diagnosis confirmation of ocular tumors.  Current Magnetic Resonance Imaging (MRI) techniques allow to image the eyes with a high level of detail and very good contrast between tissues. Consequently, having a robust and accurate segmentation tool for the Eye MRI would be a great advantage for clinicians and ophthalmologist. We have developed a framework based on Statistical Shape Models (SSM) able to fully automatically segment healthy eyes on MRIs on children. You can download the paper in the download section!
  • Automatic segmentation of ocular tumors [Published]:
Picture
a-b) 2 Examples of children eyes with ocular tumors (T1w VIBE and T2w) c) Details of different eye regions and endophytic retinoblastoma tumor
Retinoblastoma and uveal melanoma are fast spreading eye tumors usually diagnosed by using 2D Fundus Image Photography (Fundus) and 2D Ultrasound (US). Diagnosis and treatment planning of such diseases often require additional complementary imaging to confirm the tumor extend via 3D Magnetic Resonance Imaging (MRI).
​In this context, having automatic segmentations to estimate the size and the distribution of the pathological tissue would be advantageous towards tumor characterization. 
In this work, we propose a new framework for the automatic segmentation of eye structures and ocular tumors in multi-sequence MRI. ​
  • Automatic AMD identification in OCT [Public]:
​OCT is used to detect retinal diseases such as AMD. Visual inspection of OCT volumes is the main method for classification, but doing so is time consuming as each cross-section within the scan must be inspected individually. Our approach uses a novel CNN architecture to classify volume more effectively. It involves first learning a cross-section pathology classifier using potentially corrupted pseudo-labels and then using these towards a more accurate volume-level classification. Journal Article

Older Projects

  • Optimization of the geodesic active fields (GAF) for Image registration on the sphere. (MSc).