Image segmentation methods for automatic detection of the anatomical structure of the eye in people with diabetic retinopathy

• Autores: Jose Escorcia Gutierrez
• Directores de la Tesis: Aïda Valls Mateu (dir. tes.) , Pere Romero Aroca (dir. tes.) , Domènec Puig Valls (dir. tes.)
• Lectura: En la Universitat Rovira i Virgili ( España ) en 2021
• Idioma: español
• Tribunal Calificador de la Tesis: Manuel Puig Domingo (presid.) , Mohamed Abdel-Nasser (secret.) , Said David Pertuz Arroyo (voc.)
• Enlaces
  • Tesis en acceso abierto en: TDX
• Resumen

  • This thesis is framed within the comprehensive plan for early prevention of Diabetic Retinopathy (DR) launched by the Spain government following the World Health Organization to promote initiatives that raise awareness of the importance of regular eye exams among people with dia betes. Since most of the visual impairment of DR is avoidable through early detection and timely trealment, the system under investigation is projected to be a Computer-Aided Diagnosis to assist trained eye-care personnel (e.g., ophthalmologists or optometrists) through retinal imaging interpretation.

    Diabetic retinopathy has become a major worldwide threat due to an increase in the number of blind diabetic people at younger ages. The detection of DR pathologies, such as microaneurysms, hemorrhages, and exudates, through advanced computational techniques is of utmost importance in patient health care. New computer vision techniques are needed to outperform traditional screenings of color fundus images. Segmentation of the retina's whole anatomical structure is a crucial phase in detecting these pathologies.

    This thesis proposes a novel framework for fast and fully automatic optic disc segmentation based on the Modern Portfolio Theory of Markowitz to generate an innovative color fusion model capable of admitting any segmentation methodology in the medical imaging field (e.g., brain, lungs, breast, etc.). This approach acts as a powerful and real-time pre-processing stage that could be integrated into daily clinical practice to accelerate the diagnosis of DR due to its simplicity, performance, and speed. In the end, two different segmentation methods (Graph Cuts and Extended Minima + Hough Transform) are applied for optic disc detection and are compared with state-of-the-art computer vision techniques to analyze the model performance.

    The second contribution of this thesis is a method to simultaneously make a blood vessel segmentation and foveal avascular zone detection, considerably reducing the required image processing time. The pre-processing stage involves algorithms that enhance image contrast and eliminate noise artifacts. Afterward, the color spaces and their intrinsic components are examined to identify the most suitable color model to reveal the foreground pixels against the entire background. The first component of the xyY color space that represents the chrominance values derived from the XYZ color space is the most supported according to the approach developed in this thesis for blood vessel segmentation and fovea detection. This color component is the first time-related in the literature with outstanding performance compared to state-of-art results. Finally, several samples are collected for a color interpolation procedure based on statistic color information and are used by the well-known Convexity Shape Prior segmentation algorithm.

    The thesis also proposes another blood vessel segmentation method that relies on an effective feature selection based on decision tree learning. This method is validated using three different classification techniques (i.e., Decision Tree, Artificial Neural Network, and Support Vector Machine). The key point is that the feature vector is computed from only one pre-processed image in the pixel neighborhood under consideration. Subsequently, a post-processing stage is adequate for filling pixel gaps, and the removal of isolated pixels avoids the false detection of blood vessels.

    To validate the studied methods (i.e., optic disc detection and segmentation, blood vessel segmentation, and foveal avascular zone location), five image datasets are used. Four of the datasets are publicly available (i.e., DRIVE, STARE, Messidor, and HRF) and are used by many other authors, enabling the performance comparison. In addition, a new dataset has been built with images acquired from Hospital Universitari Sant Joan de Reus (Spain) to prove the applicability of our techniques to the population of our area of study.