Mariano Rincón

Identification of the optic nerve head with genetic algorithms

OBJECTIVE This work proposes creating an automatic system to locate and segment the optic nerve head (ONH) in eye fundus photographic images using genetic algorithms. METHODS AND MATERIAL Domain knowledge is used to create a set of heuristics that guide the various steps involved in the process. Initially, using an eye fundus colour image as input, a set of hypothesis points was obtained that exhibited geometric properties and intensity levels similar to the ONH contour pixels. Next, a genetic algorithm was used to find an ellipse containing the maximum number of hypothesis points in an offset of its perimeter, considering some constraints. The ellipse thus obtained is the approximation to the ONH. The segmentation method is tested in a sample of 110 eye fundus images, belonging to 55 patients with glaucoma (23.1%) and eye hypertension (76.9%) and random selected from an eye fundus image base belonging to the Ophthalmology Service at Miguel Servet Hospital, Saragossa (Spain). RESULTS AND CONCLUSIONS The results obtained are competitive with those in the literature. The methods generalization capability is reinforced when it is applied to a different image base from the one used in our study and a discrepancy curve is obtained very similar to the one obtained in our image base. In addition, the robustness of the method proposed can be seen in the high percentage of images obtained with a discrepancy delta<5 (96% and 99% in our and a different image base, respectively). The results also confirm the hypothesis that the ONH contour can be properly approached with a non-deformable ellipse. Another important aspect of the method is that it directly provides the parameters characterising the shape of the papilla: lengths of its major and minor axes, its centre of location and its orientation with regard to the horizontal position.

On the correspondence between objects and events for the diagnosis of situations in visual surveillance tasks

A key problem in visual surveillance systems (VSS) is to find an effective procedure for linking the geometric descriptions of a scene at the object level with the corresponding descriptions of the agents intervening in this scene at the activity level. In this work, we explore a constructivist approach based on using the usual Artificial Intelligence (AI) techniques and methods to establish correspondences between the entities and relations of the ontologies in these two levels. The proposal is exemplified using a real interior scenario that uses images from just one fixed camera and where the purpose of the surveillance is to do a preventive diagnosis of the activity of abandoning a potentially dangerous object in a sensitive area. The work stresses: (1) anchoring the object-level labels in the result of analytical processes on blobs, (2) specifying contextual knowledge that has to be injected to link the activities, as described by a human surveillance expert, with the objects, as they are labelled by the same expert from geometric descriptions. The work is set within the context of the 50th anniversary of AI and the leading theories on human visual perception.

A Multi-robot Surveillance System Simulated in Gazebo

A special kind of surveillance problem is the monitoring of wide enclosed areas with difficult access and changing environments. The characteristics of this kind of problem recommend a solution based on a multi-agent system, where several robots cooperate to solve the problem. A multi-robot system for surveillance in these kinds of environments has been designed and simulated on the Gazebo 3D simulator. Typical surveillance tasks are simulated and experimental results are shown.

On the effect of feedback in multilevel representation spaces for visual surveillance tasks

In this work we propose a general top-down feedback scheme between adjacent description levels to interpret video sequences. This scheme distinguishes two types of feedback: repair-oriented feedback and focus-oriented feedback. With the first it is possible to improve the systems performance and produce more reliable and consistent information, and with the second it is possible to adjust the computational load to match the aims. Finally, the general feedback scheme is used in different examples for a visual surveillance application which improved the final result of each description level by using the information in the higher adjacent level.

Knowledge modeling for the image understanding task as a design task

This paper formally analyzes the image understanding (IU) task at the knowledge level and in the observer domain. The analysis is done at three levels: task, method and domain knowledge, distinguishing the generic components in most of the IU tasks, thereby enabling the components to be reused. We model the IU problem as a design task and define a generic problem solving method (PSM) that allows us to tackle the task in a hierarchical and recursive way. The main advantage of this generic PSM is the possibility of instantiating specific PSMs through parameter space configuration, which enables the structure to be reused both across the task decomposition at different hierarchical levels and across different application domains. This work has been done following the well-established foundations of knowledge engineering that prescribe the maintenance of the conceptual structure from the modeling stage at the knowledge level down to the implementation. Finally, we apply the proposed framework to the problem of identifying the papilla in eye fundus images in order to exemplify the successive stages in the modeling process and system design, and accordingly justify the frameworks validity.

Hippocampal complex atrophy in poststroke and mild cognitive impairment

To investigate putative interacting or distinct pathways for hippocampal complex substructure (HCS) atrophy and cognitive affection in early-stage Alzheimers disease (AD) and cerebrovascular disease (CVD), we recruited healthy controls, patients with mild cognitive impairment (MCI) and poststroke patients. HCSs were segmented, and quantitative white-matter hyperintensity (WMH) load and cerebrospinal fluid (CSF) amyloid-β concentrations were determined. The WMH load was higher poststroke. All examined HCSs were smaller in amyloid-positive MCI than in controls, and the subicular regions were smaller poststroke. Memory was reduced in amyloid-positive MCI, and psychomotor speed and executive function were reduced in poststroke and amyloid-positive MCI. Size of several HCS correlated with WMH load poststroke and with CSF amyloid-β concentrations in MCI. In poststroke and amyloid-positive MCI, neuropsychological function correlated with WMH load and hippocampal volume. There are similar patterns of HCS atrophy in CVD and early-stage AD, but different HCS associations with WMH and CSF biomarkers. WMHs add to hippocampal atrophy and the archetypal AD deficit delayed recall. In line with mounting evidence of a mechanistic link between primary AD pathology and CVD, these additive effects suggest interacting pathologic processes.

Segmentation of Moving Objects with Information Feedback Between Description Levels

In real sequences, one of the factors that most negatively affects the segmentation process result is the existence of scene noise. This impairs object segmentation which has to be corrected if we wish to have some minimum guarantees of success in the following tracking or classification stages. In this work we propose a generic knowledge-based model to improve the segmentation process. Specifically, the model uses a decomposition strategy in description levels to enable the feedback of information between adjacent levels. Finally, two case studies are proposed that instantiate the model proposed for detecting humans.

Knowledge modeling for the traffic sign recognition task

In this paper we analyse the problem of traffic sign recognition at the knowledge level. Due to the complexity of the task, our approach decomposes it into simpler subtasks until the primitive level is reached. The task has been modeled at the knowledge level as a hierarchical classification task. This has allowed to discover a simple and robust Problem Solving Method (PSM) for the classification task which is reused along different classification stages of the process. The resulting system is divided into three main subtasks: image segmentation according to color, classification of the geometry of the candidate blobs and identification of the specific type of traffic sign. Finally, the system has been evaluated and the results are presented.

An annotation tool for video understanding

The interest for developing annotation tools for interpretation of video sequences arises on the own necessity of conceptualizing scenes with a suitable degree of semantics for the application domain. In this paper we analyze the features that must be present in a video annotation tool for video understanding: the entities of interest that appear in each description level of the scene are analyzed and primary features (those that must be annotated initially) and derived features (those obtained automatically) are distinguished. Lastly, we present a video annotation tool based on the previous analysis, for which the design we have chosen is modular, reusable and user friendly.

Automatic Drawing Analysis of Figures Included in Neuropsychological Tests for the Assessment and Diagnosis of Mild Cognitive Impairment

This proposal is framed within the group’s general working line of applying artificial intelligence techniques to advance in early mild cognitive impairment diagnosis. If impairment in semantic production was studied in previous works, now we rely on the reduced ability to reproduce or copy simple figures, part of standardized neuropsychological tests designed to assess mild cognitive impairment. Although the long-term goal of this project is to work with all figures from these tests, in this paper we will focus on the automatic analysis of the alternating graphs figure. We develop a quantitative descrition of different features that appear to be very abstract in the test norms and define new features that are not considered so far. Results with just one figure are quite promising (77.7% precision and 77.1 recall).