Automated image tagging through tag propagation

Trabalho apresentado no âmbito do Mestrado em Engenharia Informática, como requisito parcial Para obtenção do grau de Mestre em Engenharia Informática

Classification and staging of chronic liver disease from multimodal data

Chronic liver disease (CLD) is most of the time an asymptomatic, progressive, and ultimately potentially fatal disease. In this study, an automatic hierarchical procedure to stage CLD using ultrasound images, laboratory tests, and clinical records are described. The first stage of the proposed method, called clinical based classifier (CBC), discriminates healthy from pathologic conditions. When nonhealthy conditions are detected, the method refines the results in three exclusive pathologies in a hierarchical basis: 1) chronic hepatitis; 2) compensated cirrhosis; and 3) decompensated cirrhosis. The features used as well as the classifiers (Bayes, Parzen, support vector machine, and k-nearest neighbor) are optimally selected for each stage. A large multimodal feature database was specifically built for this study containing 30 chronic hepatitis cases, 34 compensated cirrhosis cases, and 36 decompensated cirrhosis cases, all validated after histopathologic analysis by liver biopsy. The CBC classification scheme outperformed the nonhierachical one against all scheme, achieving an overall accuracy of 98.67% for the normal detector, 87.45% for the chronic hepatitis detector, and 95.71% for the cirrhosis detector.

Enhancing Multimodal Silent Speech Interfaces With Feature Selection

In research on Silent Speech Interfaces (SSI), different sources of information (modalities) have been combined, aiming at obtaining better performance than the individual modalities. However, when combining these modalities, the dimensionality of the feature space rapidly increases, yielding the well-known "curse of dimensionality". As a consequence, in order to extract useful information from this data, one has to resort to feature selection (FS) techniques to lower the dimensionality of the learning space. In this paper, we assess the impact of FS techniques for silent speech data, in a dataset with 4 non-invasive and promising modalities, namely: video, depth, ultrasonic Doppler sensing, and surface electromyography. We consider two supervised (mutual information and Fisher's ratio) and two unsupervised (meanmedian and arithmetic mean geometric mean) FS filters. The evaluation was made by assessing the classification accuracy (word recognition error) of three well-known classifiers (knearest neighbors, support vector machines, and dynamic time warping). The key results of this study show that both unsupervised and supervised FS techniques improve on the classification accuracy on both individual and combined modalities. For instance, on the video component, we attain relative performance gains of 36.2% in error rates. FS is also useful as pre-processing for feature fusion. Copyright © 2014 ISCA.

The Acute STroke Registry and Analysis of Lausanne (ASTRAL): design and baseline analysis of an ischemic stroke registry including acute multimodal imaging.

BACKGROUND AND PURPOSE: Stroke registries are valuable tools for obtaining information about stroke epidemiology and management. The Acute STroke Registry and Analysis of Lausanne (ASTRAL) prospectively collects epidemiological, clinical, laboratory and multimodal brain imaging data of acute ischemic stroke patients in the Centre Hospitalier Universitaire Vaudois (CHUV). Here, we provide design and methods used to create ASTRAL and present baseline data of our patients (2003 to 2008). METHODS: All consecutive patients admitted to CHUV between January 1, 2003 and December 31, 2008 with acute ischemic stroke within 24 hours of symptom onset were included in ASTRAL. Patients arriving beyond 24 hours, with transient ischemic attack, intracerebral hemorrhage, subarachnoidal hemorrhage, or cerebral sinus venous thrombosis, were excluded. Recurrent ischemic strokes were registered as new events. RESULTS: Between 2003 and 2008, 1633 patients and 1742 events were registered in ASTRAL. There was a preponderance of males, even in the elderly. Cardioembolic stroke was the most frequent type of stroke. Most strokes were of minor severity (National Institute of Health Stroke Scale [NIHSS] score ≤ 4 in 40.8% of patients). Cardioembolic stroke and dissections presented with the most severe clinical picture. There was a significant number of patients with unknown onset stroke, including wake-up stroke (n=568, 33.1%). Median time from last-well time to hospital arrival was 142 minutes for known onset and 759 minutes for unknown-onset stroke. The rate of intravenous or intraarterial thrombolysis between 2003 and 2008 increased from 10.8% to 20.8% in patients admitted within 24 hours of last-well time. Acute brain imaging was performed in 1695 patients (97.3%) within 24 hours. In 1358 patients (78%) who underwent acute computed tomography angiography, 717 patients (52.8%) had significant abnormalities. Of the 1068 supratentorial stroke patients who underwent acute perfusion computed tomography (61.3%), focal hypoperfusion was demonstrated in 786 patients (73.6%). CONCLUSIONS: This hospital-based prospective registry of consecutive acute ischemic strokes incorporates demographic, clinical, metabolic, acute perfusion, and arterial imaging. It is characterized by a high proportion of minor and unknown-onset strokes, short onset-to-admission time for known-onset patients, rapidly increasing thrombolysis rates, and significant vascular and perfusion imaging abnormalities in the majority of patients.

Classification of brain tumor extracts by high resolution ¹H MRS using partial least squares discriminant analysis

High resolution proton nuclear magnetic resonance spectroscopy (¹H MRS) can be used to detect biochemical changes in vitro caused by distinct pathologies. It can reveal distinct metabolic profiles of brain tumors although the accurate analysis and classification of different spectra remains a challenge. In this study, the pattern recognition method partial least squares discriminant analysis (PLS-DA) was used to classify 11.7 T ¹H MRS spectra of brain tissue extracts from patients with brain tumors into four classes (high-grade neuroglial, low-grade neuroglial, non-neuroglial, and metastasis) and a group of control brain tissue. PLS-DA revealed 9 metabolites as the most important in group differentiation: γ-aminobutyric acid, acetoacetate, alanine, creatine, glutamate/glutamine, glycine, myo-inositol, N-acetylaspartate, and choline compounds. Leave-one-out cross-validation showed that PLS-DA was efficient in group characterization. The metabolic patterns detected can be explained on the basis of previous multimodal studies of tumor metabolism and are consistent with neoplastic cell abnormalities possibly related to high turnover, resistance to apoptosis, osmotic stress and tumor tendency to use alternative energetic pathways such as glycolysis and ketogenesis.

Multimodal semantic-associative collateral labelling and indexing of still images

A novel framework for multimodal semantic-associative collateral image labelling, aiming at associating image regions with textual keywords, is described. Both the primary image and collateral textual modalities are exploited in a cooperative and complementary fashion. The collateral content and context based knowledge is used to bias the mapping from the low-level region-based visual primitives to the high-level visual concepts defined in a visual vocabulary. We introduce the notion of collateral context, which is represented as a co-occurrence matrix, of the visual keywords, A collaborative mapping scheme is devised using statistical methods like Gaussian distribution or Euclidean distance together with collateral content and context-driven inference mechanism. Finally, we use Self Organising Maps to examine the classification and retrieval effectiveness of the proposed high-level image feature vector model which is constructed based on the image labelling results.

Semantic-associative visual content labelling and retrieval: A multimodal approach

A novel framework referred to as collaterally confirmed labelling (CCL) is proposed, aiming at localising the visual semantics to regions of interest in images with textual keywords. Both the primary image and collateral textual modalities are exploited in a mutually co-referencing and complementary fashion. The collateral content and context-based knowledge is used to bias the mapping from the low-level region-based visual primitives to the high-level visual concepts defined in a visual vocabulary. We introduce the notion of collateral context, which is represented as a co-occurrence matrix of the visual keywords. A collaborative mapping scheme is devised using statistical methods like Gaussian distribution or Euclidean distance together with collateral content and context-driven inference mechanism. We introduce a novel high-level visual content descriptor that is devised for performing semantic-based image classification and retrieval. The proposed image feature vector model is fundamentally underpinned by the CCL framework. Two different high-level image feature vector models are developed based on the CCL labelling of results for the purposes of image data clustering and retrieval, respectively. A subset of the Corel image collection has been used for evaluating our proposed method. The experimental results to-date already indicate that the proposed semantic-based visual content descriptors outperform both traditional visual and textual image feature models. (C) 2007 Elsevier B.V. All rights reserved.

Hierarchical segmentation-assisted multimodal registration for MR brain images

Information theory-based metric such as mutual information (MI) is widely used as similarity measurement for multimodal registration. Nevertheless, this metric may lead to matching ambiguity for non-rigid registration. Moreover, maximization of MI alone does not necessarily produce an optimal solution. In this paper, we propose a segmentation-assisted similarity metric based on point-wise mutual information (PMI). This similarity metric, termed SPMI, enhances the registration accuracy by considering tissue classification probabilities as prior information, which is generated from an expectation maximization (EM) algorithm. Diffeomorphic demons is then adopted as the registration model and is optimized in a hierarchical framework (H-SPMI) based on different levels of anatomical structure as prior knowledge. The proposed method is evaluated using Brainweb synthetic data and clinical fMRI images. Both qualitative and quantitative assessment were performed as well as a sensitivity analysis to the segmentation error. Compared to the pure intensity-based approaches which only maximize mutual information, we show that the proposed algorithm provides significantly better accuracy on both synthetic and clinical data.

DAEDALUS at ImageCLEF Medical Retrieval 2011: Textual, Visual and Multimodal Experiments

This paper describes the participation of DAEDALUS at ImageCLEF 2011 Medical Retrieval task. We have focused on multimodal (or mixed) experiments that combine textual and visual retrieval. The main objective of our research has been to evaluate the effect on the medical retrieval process of the existence of an extended corpus that is annotated with the image type, associated to both the image itself and also to its textual description. For this purpose, an image classifier has been developed to tag each document with its class (1st level of the hierarchy: Radiology, Microscopy, Photograph, Graphic, Other) and subclass (2nd level: AN, CT, MR, etc.). For the textual-based experiments, several runs using different semantic expansion techniques have been performed. For the visual-based retrieval, different runs are defined by the corpus used in the retrieval process and the strategy for obtaining the class and/or subclass. The best results are achieved in runs that make use of the image subclass based on the classification of the sample images. Although different multimodal strategies have been submitted, none of them has shown to be able to provide results that are at least comparable to the ones achieved by the textual retrieval alone. We believe that we have been unable to find a metric for the assessment of the relevance of the results provided by the visual and textual processes

A multimodal genre recognition prototype

In this paper, a multimodal and interactive prototype to perform music genre classification is presented. The system is oriented to multi-part files in symbolic format but it can be adapted using a transcription system to transform audio content in music scores. This prototype uses different sources of information to give a possible answer to the user. It has been developed to allow a human expert to interact with the system to improve its results. In its current implementation, it offers a limited range of interaction and multimodality. Further development aimed at full interactivity and multimodal interactions is discussed.

Deep Head Pose: Gaze-Direction Estimation in Multimodal Video

In this paper we present a convolutional neuralnetwork (CNN)-based model for human head pose estimation inlow-resolution multi-modal RGB-D data. We pose the problemas one of classification of human gazing direction. We furtherfine-tune a regressor based on the learned deep classifier. Next wecombine the two models (classification and regression) to estimateapproximate regression confidence. We present state-of-the-artresults in datasets that span the range of high-resolution humanrobot interaction (close up faces plus depth information) data tochallenging low resolution outdoor surveillance data. We buildupon our robust head-pose estimation and further introduce anew visual attention model to recover interaction with theenvironment. Using this probabilistic model, we show thatmany higher level scene understanding like human-human/sceneinteraction detection can be achieved. Our solution runs inreal-time on commercial hardware

Investigating interpreter-mediated interaction in the Chinese bilingual courtroom: Analysis based on a multimodal corpus

This research project is based on the Multimodal Corpus of Chinese Court Interpreting (MUCCCI [mutʃɪ]), a small-scale multimodal corpus on the basis of eight authentic court hearings with Chinese-English interpreting in Mainland China. The corpus has approximately 92,500 word tokens in total. Besides the transcription of linguistic and para-linguistic features, utilizing the facial expression classification rules suggested by Black and Yacoob (1995), MUCCCI also includes approximately 1,200 annotations of facial expressions linked to the six basic types of human emotions, namely, anger, disgust, happiness, surprise, sadness, and fear (Black & Yacoob, 1995). This thesis is an example of conducting qualitative analysis on interpreter-mediated courtroom interactions through a multimodal corpus. In particular, miscommunication events (MEs) and the reasons behind them were investigated in detail. During the analysis, although queries were conducted based on non-verbal annotations when searching for MEs, both verbal and non-verbal features were considered indispensable parts contributing to the entire context. This thesis also includes a detailed description of the compilation process of MUCCCI utilizing ELAN, from data collection to transcription, POS tagging and non-verbal annotation. The research aims at assessing the possibility and feasibility of conducting qualitative analysis through a multimodal corpus of court interpreting. The concept of integrating both verbal and non-verbal features to contribute to the entire context is emphasized. The qualitative analysis focusing on MEs can provide an inspiration for improving court interpreters’ performances. All the constraints and difficulties presented can be regarded as a reference for similar research in the future.

Phylogenetics, Ancestral State Reconstruction, And A New Infrafamilial Classification Of The Pantropical Ochnaceae (medusagynaceae, Ochnaceae S.str., Quiinaceae) Based On Five Dna Regions.

Ochnaceae s.str. (Malpighiales) are a pantropical family of about 500 species and 27 genera of almost exclusively woody plants. Infrafamilial classification and relationships have been controversial partially due to the lack of a robust phylogenetic framework. Including all genera except Indosinia and Perissocarpa and DNA sequence data for five DNA regions (ITS, matK, ndhF, rbcL, trnL-F), we provide for the first time a nearly complete molecular phylogenetic analysis of Ochnaceae s.l. resolving most of the phylogenetic backbone of the family. Based on this, we present a new classification of Ochnaceae s.l., with Medusagynoideae and Quiinoideae included as subfamilies and the former subfamilies Ochnoideae and Sauvagesioideae recognized at the rank of tribe. Our data support a monophyletic Ochneae, but Sauvagesieae in the traditional circumscription is paraphyletic because Testulea emerges as sister to the rest of Ochnoideae, and the next clade shows Luxemburgia+Philacra as sister group to the remaining Ochnoideae. To avoid paraphyly, we classify Luxemburgieae and Testuleeae as new tribes. The African genus Lophira, which has switched between subfamilies (here tribes) in past classifications, emerges as sister to all other Ochneae. Thus, endosperm-free seeds and ovules with partly to completely united integuments (resulting in an apparently single integument) are characters that unite all members of that tribe. The relationships within its largest clade, Ochnineae (former Ochneae), are poorly resolved, but former Ochninae (Brackenridgea, Ochna) are polyphyletic. Within Sauvagesieae, the genus Sauvagesia in its broad circumscription is polyphyletic as Sauvagesia serrata is sister to a clade of Adenarake, Sauvagesia spp., and three other genera. Within Quiinoideae, in contrast to former phylogenetic hypotheses, Lacunaria and Touroulia form a clade that is sister to Quiina. Bayesian ancestral state reconstructions showed that zygomorphic flowers with adaptations to buzz-pollination (poricidal anthers), a syncarpous gynoecium (a near-apocarpous gynoecium evolved independently in Quiinoideae and Ochninae), numerous ovules, septicidal capsules, and winged seeds with endosperm are the ancestral condition in Ochnoideae. Although in some lineages poricidal anthers were lost secondarily, the evolution of poricidal superstructures secured the maintenance of buzz-pollination in some of these genera, indicating a strong selective pressure on keeping that specialized pollination system.

Advancing Bag-of-visual-words Representations For Lesion Classification In Retinal Images.

Diabetic Retinopathy (DR) is a complication of diabetes that can lead to blindness if not readily discovered. Automated screening algorithms have the potential to improve identification of patients who need further medical attention. However, the identification of lesions must be accurate to be useful for clinical application. The bag-of-visual-words (BoVW) algorithm employs a maximum-margin classifier in a flexible framework that is able to detect the most common DR-related lesions such as microaneurysms, cotton-wool spots and hard exudates. BoVW allows to bypass the need for pre- and post-processing of the retinographic images, as well as the need of specific ad hoc techniques for identification of each type of lesion. An extensive evaluation of the BoVW model, using three large retinograph datasets (DR1, DR2 and Messidor) with different resolution and collected by different healthcare personnel, was performed. The results demonstrate that the BoVW classification approach can identify different lesions within an image without having to utilize different algorithms for each lesion reducing processing time and providing a more flexible diagnostic system. Our BoVW scheme is based on sparse low-level feature detection with a Speeded-Up Robust Features (SURF) local descriptor, and mid-level features based on semi-soft coding with max pooling. The best BoVW representation for retinal image classification was an area under the receiver operating characteristic curve (AUC-ROC) of 97.8% (exudates) and 93.5% (red lesions), applying a cross-dataset validation protocol. To assess the accuracy for detecting cases that require referral within one year, the sparse extraction technique associated with semi-soft coding and max pooling obtained an AUC of 94.2 ± 2.0%, outperforming current methods. Those results indicate that, for retinal image classification tasks in clinical practice, BoVW is equal and, in some instances, surpasses results obtained using dense detection (widely believed to be the best choice in many vision problems) for the low-level descriptors.

Hippocampal Dysplasia With Balloon Cells: Case Report And Discussion On Classification.

261