Daniel Duarte Abdala

Image segmentation fusion using general ensemble clustering methods

A new framework for adapting common ensemble clustering methods to solve the image segmentation combination problem is presented. The framework is applied to the parameter selection problem in image segmentation and compared with supervised parameter learning. We quantitatively evaluate 9 ensemble clustering methods requiring a known number of clusters and 4 with adaptive estimation of the number of clusters. Experimental results explore the capabilities of the proposed framework. It is shown that the ensemble clustering approach yields results close to the supervised learning, but without any ground truth information.

Color image segmentation guided by a color gradient network

Existing region-growing segmentation algorithms are mainly based on a static similarity concept, where only homogeneity of pixels or textures within a region plays a role. Typical natural scenes, however, show strong continuous variations of color, presenting a different, dynamic order that is not captured by existing algorithms which will segment a sky with different intensities and hues of blues or an irregularly illuminated surface as a set of different regions. We present and validate empirically a new, extremely simple approach that shows very satisfying results when applied on such scenes, while not showing poorer performance than traditional methods when applied to standard region-growing problems.

Color image segmentation using an enhanced Gradient Network Method

The objective of this paper is to evaluate a new combined approach intended for reliable color image segmentation, in particular images presenting color structures with strong but continuous color or luminosity changes, such as commonly found in outdoors scenes. The approach combines an enhanced version of the Gradient Network 2, with common region-growing approaches used as pre-segmentation steps. The GNM2 is an post-segmentation procedure based on graph analysis of global color and luminosity gradients in conjunction with a segmentation algorithm to produce a reliable segmentation result. The approach was automatically evaluated using a close/open world approach. Two different region-growing segmentation methods, CSC and Mumford and Shah with and without the GNM post-processing were compared against ground truth images using segmentation evaluation indices Rand and Bipartite Graph Matching. These results were also confronted with other well established segmentation methods (RHSEG, Watershed, EDISON, JSEG and Blobworld).

A computational method for the estimation of atrophic changes in Alzheimer's disease and mild cognitive impairment

Early diagnosis and objective monitoring of disease progression are essential for the development of therapeutic strategies in Alzheimers disease (AD). Current techniques are mainly based on semi-objective measures such as neuropsychological tests and a physicians magnetic resonance imaging (MRI) inspection. We have developed a computational method for automatic and unbiased assessment of the brains state of atrophy from MRI. Sixty-eight high-resolution MRI scans were acquired from 25 AD patients (age: 69.8+/-7.5), 16 mild cognitive impairment (MCI) patients (67.6+/-9.1) and 27 control subjects (64.9+/-8.8). On the basis of the computations we were able to recognize MCI subjects with a sensitivity of 81% and a specificity of 80% in a group of MCIs and controls using a linear classifier. To date, comparable results have only been received by manual labelling or human inspection. Our calculations are light weighted and can be applied on usual workstations in everyday clinical practice. Each step can be understood and applied by the physicians, independent of their computer knowledge. The applied image analysis process produces visual maps of atrophic changes as intermediate steps of the calculations. These can be helpful for the physician during inspection of the brain. The proposed analysis has the potential to improve AD diagnosis and treatment, especially in early its stages, and could also be used to monitor disease progression in therapeutic trials.

Ensemble Clustering via Random Walker Consensus Strategy

In this paper we present the adaptation of a random walker algorithm for combination of image segmentations to work with clustering problems. In order to achieve it, we pre-process the ensemble of clusterings to generate its graph representation. We show experimentally that a very small neighborhood will produce similar results if compared with larger choices. This fact alone improves the computational time needed to produce the final consensual clustering. We also present an experimental comparison between our results against other graph based and well known combination clustering methods in order to assess the quality of this approach.

Recording and Reenactment of Collaborative Diagnosis Sessions using DICOM

This paper presents a radiological collaborative tool capable of direct manipulation of Digital Imaging and Communications in Medicine (DICOM) images on both sides, and also recording and reenacting of a recorded session. A special collaborative application protocol formerly developed was extended and used as basis for the development of collaborative session recording and playback processes. The protocol is used today for real-time radiological meetings through the Internet. This new standard for collaborative sessions makes possible other uses for the protocol, such as asynchronous collaborative sessions, decision regulation, auditing, and educational applications. Experimental results are given which compare this protocol with other popular collaborative approaches. Comparison of these results shows that the proposed protocol performs much better than other approaches when run under controlled conditions.

Software system for calibrating examiners in cephalometric point identification.

INTRODUCTION The literature has shown that subjective concepts lead to interobserver variations in the definitions and identifications of cephalometric landmarks. Observers must be trained and calibrated to conduct scientific research using cephalometric comparisons. In this study, we aimed to develop and test a computational model called Cyclops cephalometry in radiographic cephalometry training and calibration. METHODS This system uses the concepts of evaluation process managers, examiners, and testers, thus affording uniformity in cephalometric evaluations. The system was tested with 5 orthodontists and 5 postgraduate students who located 28 landmarks in 10 lateral cephalometric radiographs before and after training. RESULTS Before training, the Student t test showed significant differences (P <0.05) in accuracy between the orthodontists and the students (71.4% and 54.9%, respectively). However, considerable improvement in accuracy was observed after training in both groups (86.5% and 83%, respectively), without significant differences (P = 0.30) between groups. Users of the system agreed about its usability aspects such as effectiveness, efficiency, and satisfaction. CONCLUSIONS This model was shown to be a useful and efficient tool in the calibration process, and might be helpful in various comparative cephalometric investigations.

Fast two-step segmentation of natural color scenes using hierarchical region-growing and a color-gradient network

We present evaluation results with focus on combined image and efficiency performance of the Gradient Network Method to segment color images, especially images showing outdoor scenes. A brief review of the techniques, Gradient Network Method and Color Structure Code, is also presented. Different region-growing segmentation results are compared against ground truth images using segmentation evaluation indices Rand and Bipartite Graph Matching. These results are also confronted with other well established segmentation methods (EDISON and JSEG). Our preliminary results show reasonable performance in comparison to several state-of-art segmentation techniques, while also showing very promising results comparatively in the terms of efficiency, indicating the applicability of our solution to real time problems.

Analyzing DICOM and non-DICOM Features in Content-Based Medical Image Retrieval: A Multi-Layer Approach

This paper presents a hybrid approach to perform content-based retrieval on medical image databases. It takes advantage of a pre-processed case base that is batch updated. DICOM information is used to perform pre-filtering to speed up the retrieval process and an image processing knowledge base is used to dynamically reconfigure the most appropriated image processing procedures to perform the image feature extraction. It shows that pre-filtering can speed up considerably the retrieval process and also that some image features produce very similar results what leads to future work on defining the needed digital image processing knowledge base

Learning a color distance metric for region-based image segmentation

In this paper we describe an experiment where we studied empirically the application of a learned distance metric to be used as discrimination function for an established color image segmentation algorithm. For this purpose we chose the Mumford-Shah energy functional and the Mahalanobis distance metric. The objective was to test our approach in an objective and quantifiable way on this specific algorithm employing this particular distance model, without making generalization claims. The empirical validation of the results was performed in two experiments: one applying the resulting segmentation method on a subset of the Berkeley Image Database, an exemplar image set possessing ground-truths and validating the results against the ground-truths using two well-known inter-cluster validation methods, namely, the Rand and BGM indexes, and another experiment using images of the same context divided into training and testing set, where the distance metric is learned from the training set and then applied to segment all the images. The obtained results suggest that the use of the specified learned distance metric provides better and more robust segmentations, even if no other modification of the segmentation algorithm is performed.