Aurélio Campilho

Image Analysis and Recognition

In this paper, we investigate applying semi-supervised clustering to audio-visual emotion analysis, a complex problem that is traditionally solved using supervised methods. We propose an extension to the semi-supervised aligned cluster analysis algorithm (SSACA), a temporal clustering algorithm that incorporates pairwise constraints in the form of must-link and cannot-link. We incorporate an exhaustive constraint propagation mechanism to further improve the clustering process. To validate the proposed method, we apply it to emotion analysis on a multimodal naturalistic emotion database. Results show substantial improvements compared to the original aligned clustering analysis algorithm (ACA) and to our previously proposed semi-supervised approach.

Segmentation of retinal blood vessels by combining the detection of centerlines and morphological reconstruction

This paper presents an automated method for the segmentation of the vascular network in retinal images. The algorithm starts with the extraction of vessel centerlines, which are used as guidelines for the subsequent vessel filling phase. For this purpose, the outputs of four directional differential operators are processed in order to select connected sets of candidate points to be further classified as centerline pixels using vessel derived features. The final segmentation is obtained using an iterative region growing method that integrates the contents of several binary images resulting from vessel width dependent morphological filters. Our approach was tested on two publicly available databases and its results are compared with recently published methods. The results demonstrate that our algorithm outperforms other solutions and approximates the average accuracy of a human observer without a significant degradation of sensitivity and specificity

On combining classifiers using sum and product rules

Abstract This paper presents a comparative study of the performance of arithmetic and geometric means as rules to combine multiple classifiers. For problems with two classes, we prove that these combination rules are equivalent when using two classifiers and the sum of the estimates of the a posteriori probabilities is equal to one. We also prove that the case of a two class problem and a combination of two classifiers is the only one where such equivalence occurs. We present experiments illustrating the equivalence of the rules under the above mentioned assumptions.

The Arabidopsis D-type cyclin CYCD2;1 and the inhibitor ICK2/KRP2 modulate auxin-induced lateral root formation.

Root branching is stimulated by auxin and occurs as a result of lateral roots initiated from the pericycle. Here, new light is shed on how the cell cycle is regulated during lateral root initiation by two interacting cell cycle regulatory proteins, the D-type cyclin CYCD2;1 and the auxin-regulated inhibitory protein ICK2/KRP2. The integration of cell division in root growth and development requires mediation of developmental and physiological signals through regulation of cyclin-dependent kinase activity. Cells within the pericycle form de novo lateral root meristems, and D-type cyclins (CYCD), as regulators of the G1-to-S phase cell cycle transition, are anticipated to play a role. Here, we show that the D-type cyclin protein CYCD2;1 is nuclear in Arabidopsis thaliana root cells, with the highest concentration in apical and lateral meristems. Loss of CYCD2;1 has a marginal effect on unstimulated lateral root density, but CYCD2;1 is rate-limiting for the response to low levels of exogenous auxin. However, while CYCD2;1 expression requires sucrose, it does not respond to auxin. The protein Inhibitor-Interactor of CDK/Kip Related Protein2 (ICK2/KRP2), which interacts with CYCD2;1, inhibits lateral root formation, and ick2/krp2 mutants show increased lateral root density. ICK2/KRP2 can modulate the nuclear levels of CYCD2;1, and since auxin reduces ICK2/KRP2 protein levels, it affects both activity and cellular distribution of CYCD2;1. Hence, as ICK2/KRP2 levels decrease, the increase in lateral root density depends on CYCD2;1, irrespective of ICK2/CYCD2;1 nuclear localization. We propose that ICK2/KRP2 restrains root ramification by maintaining CYCD2;1 inactive and that this modulates pericycle responses to auxin fluctuations.

Segmentation of the carotid intima-media region in B-mode ultrasound images

This paper proposes a new approach for the segmentation of both near-end and far-end intima-media regions of the common carotid artery in ultrasound images. The method requires minimal user interaction and is able to segment the near-end wall in arteries with large, hypoechogenic and irregular plaques, issues usually not considered previously due to the increased segmentation difficulty. The adventitia is detected by searching for the best fit of a cubic spline to edges having features compatible with the adventitia boundary. The algorithm uses a global smoothness constraint and integrates discriminating features of the adventitia to reduce the attraction by other edges. Afterwards, using the information of the adventitia location, the lumen boundary is detected by combining dynamic programming, smooth intensity thresholding surfaces and geometric snakes. Smooth contours that correctly adapt to the intima are produced, even in the presence of deep concavities. Moreover, unlike balloon-based snakes, the propagation force does not depend on gradients and does not require a predefined direction. An extensive statistical evaluation is computed, using a set of 47 images from 24 different symptomatic patients, including several classes, sizes and shapes of plaques. Bland-Altman plots of the mean intima-media thickness, for manual segmentations of two medical experts, show a high intra-observer and inter-observer agreement, with mean differences close to zero (mean between -0.10mm and 0.18mm) and with the large majority of differences within the limits of agreement (standard deviation between 0.10mm and 0.12mm). Similar plots reveal a good agreement between the automatic and the manual segmentations (mean between -0.07mm and 0.11mm and standard deviation between 0.11mm and 0.12mm).

An Automatic Graph-Based Approach for Artery/Vein Classification in Retinal Images

The classification of retinal vessels into artery/vein (A/V) is an important phase for automating the detection of vascular changes, and for the calculation of characteristic signs associated with several systemic diseases such as diabetes, hypertension, and other cardiovascular conditions. This paper presents an automatic approach for A/V classification based on the analysis of a graph extracted from the retinal vasculature. The proposed method classifies the entire vascular tree deciding on the type of each intersection point (graph nodes) and assigning one of two labels to each vessel segment (graph links). Final classification of a vessel segment as A/V is performed through the combination of the graph-based labeling results with a set of intensity features. The results of this proposed method are compared with manual labeling for three public databases. Accuracy values of 88.3%, 87.4%, and 89.8% are obtained for the images of the INSPIRE-AVR, DRIVE, and VICAVR databases, respectively. These results demonstrate that our method outperforms recent approaches for A/V classification.

Real-time implementation of an optical flow algorithm

This paper presents the implementation of an optical flow algorithm on a pipeline image processor The overall optical flow computation method is presented and evaluated on a common set of image sequences. Results are compared to other implementations according to two different error measures. Due to its deterministic architecture, this implementation achieves very low computation delays that allow it to operate at standard video frame-rate and resolutions. It compares favorably to recent implementations in parallel hardware.

Classification of breast cancer histology images using Convolutional Neural Networks

Breast cancer is one of the main causes of cancer death worldwide. The diagnosis of biopsy tissue with hematoxylin and eosin stained images is non-trivial and specialists often disagree on the final diagnosis. Computer-aided Diagnosis systems contribute to reduce the cost and increase the efficiency of this process. Conventional classification approaches rely on feature extraction methods designed for a specific problem based on field-knowledge. To overcome the many difficulties of the feature-based approaches, deep learning methods are becoming important alternatives. A method for the classification of hematoxylin and eosin stained breast biopsy images using Convolutional Neural Networks (CNNs) is proposed. Images are classified in four classes, normal tissue, benign lesion, in situ carcinoma and invasive carcinoma, and in two classes, carcinoma and non-carcinoma. The architecture of the network is designed to retrieve information at different scales, including both nuclei and overall tissue organization. This design allows the extension of the proposed system to whole-slide histology images. The features extracted by the CNN are also used for training a Support Vector Machine classifier. Accuracies of 77.8% for four class and 83.3% for carcinoma/non-carcinoma are achieved. The sensitivity of our method for cancer cases is 95.6%.

Segmentation of ultrasound images of the carotid using RANSAC and cubic splines

A new algorithm is proposed for the semi-automatic segmentation of the near-end and the far-end adventitia boundary of the common carotid artery in ultrasound images. It uses the random sample consensus method to estimate the most significant cubic splines fitting the edge map of a longitudinal section. The consensus of the geometric model (a spline) is evaluated through a new gain function, which integrates the responses to different discriminating features of the carotid boundary: the proximity of the geometric model to any edge or to valley shaped edges; the consistency between the orientation of the normal to the geometric model and the intensity gradient; and the distance to a rough estimate of the lumen boundary. A set of 50 longitudinal B-mode images of the common carotid and their manual segmentations performed by two medical experts were used to assess the performance of the method. The image set was taken from 25 different subjects, most of them having plaques of different classes (class II to class IV), sizes and shapes. The quantitative evaluation showed promising results, having detection errors similar to the ones observed in manual segmentations for 95% of the far-end boundaries and 73% of the near-end boundaries.

Optic disc segmentation using the sliding band filter

BACKGROUND The optic disc (OD) centre and boundary are important landmarks in retinal images and are essential for automating the calculation of health biomarkers related with some prevalent systemic disorders, such as diabetes, hypertension, cerebrovascular and cardiovascular diseases. METHODS This paper presents an automatic approach for OD segmentation using a multiresolution sliding band filter (SBF). After the preprocessing phase, a low-resolution SBF is applied on a downsampled retinal image and the locations of maximal filter response are used for focusing the analysis on a reduced region of interest (ROI). A high-resolution SBF is applied to obtain a set of pixels associated with the maximum response of the SBF, giving a coarse estimation of the OD boundary, which is regularized using a smoothing algorithm. RESULTS Our results are compared with manually extracted boundaries from public databases (ONHSD, MESSIDOR and INSPIRE-AVR datasets) outperforming recent approaches for OD segmentation. For the ONHSD, 44% of the results are classified as Excellent, while the remaining images are distributed between the Good (47%) and Fair (9%) categories. An average overlapping area of 83%, 89% and 85% is achieved for the images in ONHSD, MESSIDOR and INSPIR-AVR datasets, respectively, when comparing with the manually delineated OD regions. DISCUSSION The evaluation results on the images of three datasets demonstrate the better performance of the proposed method compared to recently published OD segmentation approaches and prove the independence of this method when from changes in image characteristics such as size, quality and camera field of view.