Biomedical Image Processing Group

Video Stabilization

Bruno, Paulo and Ricardo 2012

The automatic detection and tracking of leukocytes in intravital microscopy images can ensure more accurate analysis and consequently help researchers in the development of more effective therapeutic strategies. However, in an in-vivo study, the respiration and cardiac activity of the animal cause a momentary loss of focus of the vessels, causing blurred and distorted images. This fact considerably hinders the detection and tracking of leukocytes. Therefore, this work proposes a multiresolution method for the stabilization of intravital microscopy image sequence using registration techniques. The results obtained from proposed method indicate a significant visual improvement in image stabilization.

Image Restoration

Carlos and Ricardo 2013

One of the most significant challenges in the development of automatic methods for intravital microscopy is the stabilization and removal of motion artifacts from a video, caused mainly by the animal's breathing. Such movements lead to quick changes in the microscope's focal plane and, consequently, generates blur and distortions in the images. By stabilizing the intravital microscopy video images, we expect that the automatic tracking of leukocytes will become more robust and it will allow the assessment of the movements of these cells. Therefore, the primary objective of this research is the study of deconvolution techniques applied to the restoration of intravital microscopy video images corrupted by motion artifacts.

Motion Blur Detection

Carlos, Bruno and Ricardo 2013

Due to the limited control over the conditions of the image acquisition, unavoidable motion blur resulting mainly from the heartbeat and respiratory movements of the in vivo specimen will always be present. This problem can significantly undermine the results of either visual or computerized analysis of the acquired video. Since severe motion blur usually corrupts only a fraction of the number of images, it is essential and desirable to have a procedure to identify them for posterior restoration automatically. In this research, we have been investigating two different techniques for the detection of motion blur.

Detection of Leukocytes

Bruno, Paulo, Kathiani and Ricardo 2012

In recent years, a large number of researchers have focused their efforts and interests for the in vivo study of cellular and molecular mechanisms of leukocyte-endothelial interactions in the microcirculation of various tissues and various inflammatory conditions. The purpose of these studies is to develop more effective therapeutic strategies for the treatment of inflammatory and autoimmune diseases. In these types of studies, the intravital microscopy imaging is considered to be the "gold standard" technique since it allows the acquisition of images with high temporal resolution and low spatial depth. Currently, analysis of leukocyte-endothelial interactions in small animals is performed visually using intravital microscopy image sequences. Besides being time-consuming, this procedure can lead to visual fatigue of the observer and thus generate false statistics. In this context, this research line aims to study and develop computer techniques for the automatic detection and tracking of leukocytes in intravital video microscopy.

Effect of Image Registration on MS-lesions

Paulo and Ricardo 2014

Relatively recently, some automatic image processing systems have been proposed in the literature to help radiologists detect and compute the volume of MS lesions in MR images. Among most of these approaches, registration of multi-contrast MR clinical images (T1/T2/PD/FLAIR), as well as the registration of patient's data to anatomical atlases, have proved to be essential steps to successfully detect and quantitatively assess lesion load in MS in the neuroimaging processing pipeline. However, despite its importance and frequent use in automatic image processing systems, the effect of MS lesions on the final results of image registration has not been thoroughly investigated. In this work, image registration techniques using both affine and deformable transformations were used to assess if MS lesions (stratified in mild, moderate or severe) had any effect on the final results of image registration.

Segmentation of MS-lesions

Paulo, Bruno and Ricardo 2013

Nowadays, the preferred method to segment MS lesions is to delimit them in 3D MR images manually; in this case, a specialist with the limited help of a computer does the task. However, this approach is expensive and error-prone between specialists, given that the lesions edge contrast is low. The difficulties in the automatic detection and segmentation of MS lesions in MR images are their variability in size and location, their low contrast, due to partial volume effect, and the high range of forms (highlighted, not highlighted, black holes) the lesions can assume depending on the stage of the disease. Recently, many researchers have turned their efforts to develop techniques that aim to reduce the amount of time spent on image analysis and to measure in a more precise way the volume of brain tissues and MS lesions.

Detection of the midsaggital plane in MR images

Camilo, Carlos, Ricardo 2015

The midsagittal plane (MSP) separates the cerebrum into left and right hemispheres, and its detection has many useful applications in brain image processing. We propose an automatic technique to find the MSP in magnetic resonance (MR) images that uses a planar measure obtained from eigenanalysis of a local matrix of second-order moments of 3D phase congruency responses to determine those voxels most likely to belong to the MSP. A weighted least-squares fitting algorithm is used in a coarse-to-fine iterative manner to find the best plane (the MSP) through the selected voxels. Unlike most of the proposed approaches, which mainly rely on symmetric measures of the brain, our technique uses a direct measurement to find the MSP. We applied our method to 202 MR images (40 clinical and 162 synthetic images), and it has shown to work for both symmetrical and asymmetrical brain images. Quantitative assessment, using the angle between unit normals of the detected and reference MSPs, yielded to mean absolute angle value bellow 0. 5 degrees.

​

Probabilistic Atlas of 3D Salient Points

Carlos and Ricardo 2014

Usually, the structures of interest are manually segmented, which is an error-prone and time-consuming task. For this reason, many researchers have turned their efforts to the development of automatic techniques for the segmentation of tissues and brain structures in MR images. Among the approaches proposed in the literature, the ones based on geometric deformable models using probabilistic and topological atlases are among the techniques presenting the best results. The reason is that they allow the use of anatomical information inherently contained in the meshes during the segmentation process. However, a significant difficulty applying geometric deformable models for medical image segmentation is the proper initial positioning of the model. Thus, it is intended, for this research proposal, the improvement of a technique for automatic detection of 3D salient points and, from this, the development of a probabilistic atlas of salient points that will help to automate the initial positioning of deformable geometric models. Thus, segmentation techniques based on this approach may be more efficient and enable higher accuracy and speed in volumetric measurements of brain structures.