Antonio Mosquera González

Defining the Optimal Region of Interest for Hyperemia Grading in the Bulbar Conjunctiva

Conjunctival hyperemia or conjunctival redness is a symptom that can be associated with a broad group of ocular diseases. Its levels of severity are represented by standard photographic charts that are visually compared with the patients eye. This way, the hyperemia diagnosis becomes a nonrepeatable task that depends on the experience of the grader. To solve this problem, we have proposed a computer-aided methodology that comprises three main stages: the segmentation of the conjunctiva, the extraction of features in this region based on colour and the presence of blood vessels, and, finally, the transformation of these features into grading scale values by means of regression techniques. However, the conjunctival segmentation can be slightly inaccurate mainly due to illumination issues. In this work, we analyse the relevance of different features with respect to their location within the conjunctiva in order to delimit a reliable region of interest for the grading. The results show that the automatic procedure behaves like an expert using only a limited region of interest within the conjunctiva.

Application of a Multilayer Discrete-Time CNN to Deformable Models

In this work Cellular Neural Networks are applied to image analysis techniques as a deformable models. To this end the problem is considered based on a discrete-time CNN with cyclic templates and time-variant external inputs. The appropriateness for a VLSI implementation and massively parallel computing of CNNs will permit a considerable improvement in processing speed with respect to the clasical active contours approaches.

Texture Image Segmentatoin Using a Modified Hopfield Network

In this work we describe the implementation of an artificial neural network, an extension of Hopfields model, for the supervised segmentation of textured images. We use a Markov random field in order to model the textures in the image. The problem is approached in terms of the minimization of a objective function which integrates statistical and spatial information and which is projected onto the network. It provides a locally optimal solution to the problem of the classification of M*M pixels into K classes (textures). The experimental results obtained on artificial and natural images show the validity of the architecture we propose.

Precise segmentation of the bulbar conjunctiva for hyperaemia images

Hyperaemia is an excess of blood in a tissue that causes the appearance of an unusual red hue in the affected area. It is a common occurrence in the bulbar conjunctiva, where it can be related to multiple pathologies, such as conjunctivitis or dry eye syndrome. Specialists grade hyperaemia by means of a tedious, subjective, non-repeatable and time-consuming process. These drawbacks can be solved with the automatisation of the process by means of image processing techniques. The automatic segmentation of the conjunctiva is an important part of the process, as it ensures the absence of noise in posterior stages of the methodology. However, there are several issues of illumination and focus in the input videos that difficult the process. In this work, several segmentation algorithms are proposed and compared in order to obtain an accurate location of the bulbar conjunctiva.

Assessment of the repeatability in an automatic methodology for hyperemia grading in the bulbar conjunctiva

When the vessels of the bulbar conjunctiva get congested with blood, a characteristic red hue appears in the area. This symptom is known as hyperemia, and can be an early indicator of certain pathologies. Therefore, a prompt diagnosis is desirable in order to minimize both medical and economic repercussions. A fully automatic methodology for hyperemia grading in the bulbar conjunctiva was developed, by means of image processing and machine learning techniques. As there is a wide range of illumination, contrast, and focus issues in the images that specialists use to perform the grading, a repeatability analysis is necessary. Thus, the validation of each step of the methodology was performed, analyzing how variations in the images are translated to the results, and comparing them to the optometrists measurements. Our results prove the robustness of our methodology to various conditions. Moreover, the differences in the automatic outputs are similar to the optometrists ones.

Optimization Neural Networks for Image Segmentation

In this work we describe the implementation of an artificial neural network, an extension of Hopfields model, for the segmentation of images. The problem is approached in terms of the minimization of an objective function which integrates statistical and spatial information and which is projected onto the network. It provides a locally optimal solution to the problem of the classification of N1*N2 pixels into M classes. The experimental results obtained show the validity of the architecture we propose.