187

Breast cancer remains one of the most prevalent and deadly cancers worldwide, making accurate evaluation of molecular markers important for effective disease management. Biomarkers such as ER, PR, and HER2 are typically assessed because they help inform prognosis and guide treatment decisions. Predicting these characteristics from imaging can support earlier clinical intervention, reduce reliance on invasive procedures, and contribute to more personalized care. While radiomics and deep learning approaches have demonstrated potential, comprehensive comparisons across these methods are still limited. This study evaluated handcrafted features, deep features, and end-to-end deep learning models for predicting ER, PR, and HER2 status from DCE-MRI. Each feature type was first assessed individually and then combined using early and late fusion. Handcrafted and deep features were processed through a pipeline that included resampling, dimensionality reduction, and model selection, while end-to-end models were trained using different initialization strategies and loss functions. The best models achieved AUCs of 0.659 for ER, 0.679 for PR, and 0.686 for HER2. Although late fusion generally improved performance, bias toward the majority classes persisted. Overall, the results suggest that combining different modeling strategies may enhance robustness in breast cancer characterization.

Deep learning in computational pathology (CPath) has rapidly advanced in recent years. Research has primarily focused on enhancing accuracy and interpretability across various histology image analysis tasks, from tile-level to slide-level foundation models and novel multiple instance learning (MIL) strategies. However, it is equally important for models to provide well-calibrated confidence estimates. Due to factors such as dataset bias, overfitting, and limited training data, existing models tend to be overly confident on test sets. Promising solutions to address this issue include temperature scaling, a post-hoc method that adjusts logits using a single scalar value. However, the role of calibration in CPath is yet to be clarified. In this study, we evaluate temperature scaling and linear temperature scaling for CPath tasks, analyzing their impact on recalibration in both in-domain and out-of-domain distributions. The results show the limitations of current probability calibration techniques and motivate future work.

The use of Convolutional Neural Network (CNN) models for image classification tasks has gained significant popularity. However, the lack of interpretability in CNN models poses challenges for debugging and validation. To address this issue, various explanation methods have been developed to provide insights into CNN models. This paper focuses on the validity of these explanation methods for ordinal regression tasks, where the classes have a predefined order relationship. Different modifications are proposed for two explanation methods to exploit the ordinal relationships between classes: Grad-CAM based on Ordinal Binary Decomposition (GradOBD-CAM) and Ordinal Information Bottleneck Analysis (OIBA). The performance of these modified methods is compared to existing popular alternatives. Experimental results demonstrate that GradOBD-CAM outperforms other methods in terms of interpretability for three out of four datasets, while OIBA achieves superior performance compared to IBA. • Addressing interpretability challenges in CNN models for ordinal regression. • Modification of explanation methods for ordinal regression tasks. • Superior performance of GradOBD-CAM for interpretability in 3 out of 4 datasets. • Enhanced interpretability with OIBA compared to IBA. • Evaluation using comprehensive ordinal degradation score metrics.

In order to facilitate the adoption of deep learning in areas where decisions are of critical importance, understanding the model’s internal workings is paramount. Nevertheless, since most models are considered black boxes, this task is usually not trivial, especially when the user does not have access to the network’s intermediate outputs. In this paper, we propose IBISA, a model-agnostic attribution method that reaches state-of-the-art performance by optimizing sampling masks using the Information Bottleneck Principle. Our method improves on the previously known RISE and IBA techniques by placing the bottleneck right after the image input without complex formulations to estimate the mutual information. The method also requires only twenty forward passes and ten backward passes through the network, which is significantly faster than RISE, which needs at least 4000 forward passes. We evaluated IBISA using a VGG-16 and a ResNET-50 model, showing that our method produces explanations comparable or superior to IBA, RISE, and Grad-CAM but much more efficiently. • We introduce IBISA, a new model-agnostic approach to generate saliency maps. • The Information Bottleneck principle is used to optimize masks in the input. • IBISA reaches state-of-the-art performance with highly reduced computational cost.

Keratitis is an inflammatory corneal condition responsible for 10% of visual impairment in low- and middle-income countries (LMICs), with bacteria, fungi, or amoeba as the most common infection etiologies. While an accurate and timely diagnosis is crucial for the selected treatment and the patients’ sight outcomes, due to the high cost and limited availability of laboratory diagnostics in LMICs, diagnosis is often made by clinical observation alone, despite its lower accuracy. In this study, we investigate and compare different deep learning approaches to diagnose the source of infection: 1) three separate binary models for infection type predictions; 2) a multitask model with a shared backbone and three parallel classification layers (Multitask V1); and, 3) a multitask model with a shared backbone and a multi-head classification layer (Multitask V2). We used a private Brazilian cornea dataset to conduct the empirical evaluation. We achieved the best results with Multitask V2, with an area under the receiver operating characteristic curve (AUROC) confidence intervals of 0.7413-0.7740 (bacteria), 0.83950.8725 (fungi), and 0.9448-0.9616 (amoeba). A statistical analysis of the impact of patient features on models’ performance revealed that sex significantly affects amoeba infection prediction, and age seems to affect fungi and bacteria predictions.

Case-based explanations are an intuitive method to gain insight into the decision-making process of deep learning models in clinical contexts. However, medical images cannot be shared as explanations due to privacy concerns. To address this problem, we propose a novel method for disentangling identity and medical characteristics of images and apply it to anonymize medical images. The disentanglement mechanism replaces some feature vectors in an image while ensuring that the remaining features are preserved, obtaining independent feature vectors that encode the images' identity and medical characteristics. We also propose a model to manufacture synthetic privacy-preserving identities to replace the original image's identity and achieve anonymization. The models are applied to medical and biometric datasets, demonstrating their capacity to generate realistic-looking anonymized images that preserve their original medical content. Additionally, the experiments show the network's inherent capacity to generate counterfactual images through the replacement of medical features.

The development of accurate methods for OCT image analysis is highly dependent on the availability of large annotated datasets. As such datasets are usually expensive and hard to obtain, novel approaches based on deep generative models have been proposed for data augmentation. In this work, a flow-based network (SRFlow) and a generative adversarial network (ESRGAN) are used for synthesizing high-resolution OCT B-scans from low-resolution versions of real OCT images. The quality of the images generated by the two models is assessed using two standard fidelity-oriented metrics and a learned perceptual quality metric. The performance of two classification models trained on real and synthetic images is also evaluated. The obtained results show that the images generated by SRFlow preserve higher fidelity to the ground truth, while the outputs of ESRGAN present, on average, better perceptual quality. Independently of the architecture of the network chosen to classify the OCT B-scans, the model's performance always improves when images generated by SRFlow are included in the training set.

Optical Coherence Tomography (OCT) is the major diagnostic tool for the leading cause of blindness in developed countries: Diabetic Macular Edema (DME). Depending on the type of fluid accumulations, different treatments are needed. In particular, Cystoid Macular Edemas (CMEs) represent the most severe scenario, while Diffuse Retinal Thickening (DRT) is an early indicator of the disease but a challenging scenario to detect. While methodologies exist, their explanatory power is limited to the input sample itself. However, due to the complexity of these accumulations, this may not be enough for a clinician to assess the validity of the classification. Thus, in this work, we propose a novel approach based on multi-prototype networks with vision transformers to obtain an example-based explainable classification. Our proposal achieved robust results in two representative OCT devices, with a mean accuracy of 0.9099 ± 0.0083 and 0.8582 ± 0.0126 for CME and DRT-type fluid accumulations, respectively.

The application of machine learning algorithms to predict the mechanism of action (MoA) of drugs can be highly valuable and enable the discovery of new uses for known molecules. The developed methods are usually evaluated with small subsets of MoAs with large support, leading to deceptively good generalization. However, these datasets may not accurately represent a practical use, due to the limited number of target MoAs. Accurate predictions for these rare drugs are important for drug discovery and should be a point of focus. In this work, we explore different training strategies to improve the performance of a well established deep learning model for rare drug MoA prediction. We explored transfer learning by first learning a model for common MoAs, and then using it to initialize the learning of another model for rarer MoAs. We also investigated the use of a cascaded methodology, in which results from an initial model are used as additional inputs to the model for rare MoAs. Finally, we proposed and tested an extension of Mixup data augmentation for multilabel classification. The baseline model showed an AUC of 73.2% for common MoAs and 62.4% for rarer classes. From the investigated methods, Mixup alone failed to improve the performance of a baseline classifier. Nonetheless, the other proposed methods outperformed the baseline for rare classes. Transfer Learning was preferred in predicting classes with less than 10 training samples, while the cascaded classifiers (with Mixup) showed better predictions for MoAs with more than 10 samples. However, the performance for rarer MoAs still lags behind the performance for frequent MoAs and is not sufficient for the reliable prediction of rare MoAs.

This paper presents a summary of the Competition on Face Morphing Attack Detection Based on Privacy-aware Synthetic Training Data (SYN-MAD) held at the 2022 In-ternational Joint Conference on Biometrics (IJCB 2022). The competition attracted a total of 12 participating teams, both from academia and industry and present in 11 differ-ent countries. In the end, seven valid submissions were submitted by the participating teams and evaluated by the organizers. The competition was held to present and at-tract solutions that deal with detecting face morphing at-tacks while protecting people's privacy for ethical and le-gal reasons. To ensure this, the training data was limited to synthetic data provided by the organizers. The submitted solutions presented innovations that led to out-performing the considered baseline in many experimental settings. The evaluation benchmark is now available at: https://github.com/marcohuber/SYN-MAD-2022.

This work summarizes the IJCB Occluded Face Recognition Competition 2022 (IJCB-OCFR-2022) embraced by the 2022 International Joint Conference on Biometrics (IJCB 2022). OCFR-2022 attracted a total of 3 participating teams, from academia. Eventually, six valid submissions were submitted and then evaluated by the organizers. The competition was held to address the challenge of face recognition in the presence of severe face occlusions. The participants were free to use any training data and the testing data was built by the organisers by synthetically occluding parts of the face images using a well-known dataset. The submitted solutions presented innovations and performed very competitively with the considered baseline. A major output of this competition is a challenging, realistic, and diverse, and publicly available occluded face recognition benchmark with well defined evaluation protocols.

Multiple Sclerosis (MS) is a chronic and inflammatory disorder that causes degeneration of axons in brain white matter and spinal cord. Magnetic Resonance Imaging (MRI) is extensively used to identify MS lesions and evaluate the progression of the disease, but the manual identification and quantification of lesions are time consuming and error-prone tasks. Thus, automated Deep Learning methods, in special Convolutional Neural Networks (CNNs), are becoming popular to segment medical images. It has been noticed that the performance of those methods tends to decrease when applied to MRI acquired under different protocols. The aim of this work is to statistically evaluate the possible influence of domain adaptation during the training process of CNNs models for segmenting MS lesions in MRI. The segmentation models were tested on MRIs (FLAIR and T1) of 20 patients diagnosed with Multiple Sclerosis. The set of segmented images of each different model was compared statistically, through the metrics Dice Similarity Coefficient (DSC), Predictive Positive Value (PPV) and Absolute Volume Difference (AVD). The results indicate that the domain adapted training can improve the performance of automatic segmentation methods, by CNNs, and have great potential to be used in medical clinics in the future.

SARS-CoV-2 has presented direct and indirect challenges to the scientific community. One of the most prominent indirect challenges advents from the mandatory use of face masks in a large number of countries. Face recognition methods struggle to perform identity verification with similar accuracy on masked and unmasked individuals. It has been shown that the performance of these methods drops considerably in the presence of face masks, especially if the reference image is unmasked. We propose FocusFace, a multi-task architecture that uses contrastive learning to be able to accurately perform masked face recognition. The proposed architecture is designed to be trained from scratch or to work on top of state-of-the-art face recognition methods without sacrificing the capabilities of a existing models in conventional face recognition tasks. We also explore different approaches to design the contrastive learning module. Results are presented in terms of masked-masked (M-M) and unmasked-masked (U-M) face verification performance. For both settings, the results are on par with published methods, but for M-M specifically, the proposed method was able to outperform all the solutions that it was compared to. We further show that when using our method on top of already existing methods the training computational costs decrease significantly while retaining similar performances. The implementation and the trained models are available at GitHub.

This paper presents a summary of the Masked Face Recognition Competitions (MFR) held within the 2021 International Joint Conference on Biometrics (IJCB 2021). The competition attracted a total of 10 participating teams with valid submissions. The affiliations of these teams are diverse and associated with academia and industry in nine different countries. These teams successfully submitted 18 valid solutions. The competition is designed to motivate solutions aiming at enhancing the face recognition accuracy of masked faces. Moreover, the competition considered the deployability of the proposed solutions by taking the compactness of the face recognition models into account. A private dataset representing a collaborative, multisession, real masked, capture scenario is used to evaluate the submitted solutions. In comparison to one of the topperforming academic face recognition solutions, 10 out of the 18 submitted solutions did score higher masked face verification accuracy.

As a key technology to help bridging the gap between deaf and hearing people, sign language recognition (SLR) has become one of the most active research topics in the human-computer interaction field. Although several SLR methodologies have been proposed, the development of a real-world SLR system is still a very challenging task. One of the main challenges is related to the large intersigner variability that exists in the manual signing process of sign languages. To address this problem, we propose a novel end-to-end deep neural network that explicitly models highly discriminative signer-independent latent representations from the input data. The key idea of our model is to learn a distribution over latent representations, conditionally independent of signer identity. Accordingly, the learned latent representations will preserve as much information as possible about the signs, and discard signer-specific traits that are irrelevant for recognition. By imposing such regularization in the representation space, the result is a truly signer-independent model which is robust to different and new test signers. The experimental results demonstrate the effectiveness of the proposed model in several SLR databases.

Magnetic resonance imaging (MRI) is the most commonly used exam for diagnosis and follow-up of neurodegenerative diseases, such as multiple sclerosis (MS). MS is a neuroinflammatory and neurodegenerative disease characterized by demyelination of neuron axon. This demyelination process causes lesions in white matter that can be observed in vivo by MRI. Such lesions may provide quantitative assessments of the inflammatory activity of the disease. Quantitative measures based on various features of lesions have been shown to be useful in clinical trials for evaluating therapies. Although manual segmentations are considered as the gold standard, this process is time consuming and error prone. Therefore, automated lesion identification and quantification of the MRI are active areas in MS research. The purpose of this study was to perform the brain lesions volumetric quantification in MS patients, after segmentation via a convolutional neural network (CNN) model. Initially, MRI was rigidly registered, skullstripped and bias corrected. After, we use the CNN for brain lesions segmentation, which used training data to identify lesions within new test subjects. Finally, volume quantification was performed with a count of segmented voxels and represented by mm3. We did not observe a statistical difference between the volume of brain lesion automatically identified and the volume manually segmented. The use of deep learning techniques in health is constantly developing. We observed that the use of these computational method for segmentation and quantification of brain lesions can be applied to aid in diagnosis and follow-up of MS.

Convolutional neural networks are shown to be vulnerable to changes in the background. The proposed method is an end-to-end method that augments the training set by introducing new backgrounds during the training process. These backgrounds are created by a generative network that is trained as an adversary to the model. A case study is explored based on overhead power line insulators detection using a drone - a training set is prepared from photographs taken inside a laboratory and then evaluated using photographs that are harder to collect from outside the laboratory. The proposed method improves performance by over 20% for this case study.

A key to the generalization ability of Convolutional Neural Networks (CNNs) is the idea that patterns that appear in one region of the image have a high probability of appearing in other regions. This notion is also true for other spatial relationships, such as orientation. Motivated by the fact that in the early layers of CNNs distinct filters often encode for the same feature at different angles, we propose to incorporate the rotation equivariant prior in these models. In this work, different regularization strategies that capture the notion of approximate equivariance were designed and quantitatively evaluated in their ability to generate rotation-equivariant models and their effect on the model's capacity to generalize to unseen data. Some of these strategies consistently lead to higher test set accuracies when compared to a baseline model, on classification tasks. We conclude that the rotation equivariance prior should be adopted in the general setting when modeling visual data.

Deep learning has offered significant performance improvements on several pattern recognition problems. However, the well-known need for large amounts of labeled data limits applicability and performance where those are not available. Hence, this paper proposes an adaptation of the triplet loss for self-learning with entirely unlabeled data, where there is uncertainty in the generated triplets. The methodology was applied to off-the-person electrocardiogram-based biometric authentication and unconstrained face identity verification tasks, including stress experiments designed to simulate more difficult circumstances. Despite the uncertainty related to the use of unlabeled data, the method was mostly capable of avoiding negatively affecting the model's performance. The promising results show the proposed method can be a viable alternative to supervised learning in cases where only unlabeled data are available. The method is especially suitable for training with continuous stream-based datasets such as on person re-identification in video streams and continuous electrocardiogram-based biometrics.

With the appearance of Shared Autonomous Vehicles there will no longer be a driver responsible for maintaining the car interior and well-being of passengers. To counter this, it is imperative to have a system that is able to detect any abnormal behaviors, more specifically, violence between passengers. Traditional action recognition algorithms build models around known interactions but activities can be so diverse, that having a dataset that incorporates most use cases is unattainable. While action recognition models are normally trained on all the defined activities and directly output a score that classifies the likelihood of violence, video anomaly detection algorithms present themselves as an alternative approach to build a good discriminative model since usually only non-violent examples are needed. This work focuses on anomaly detection and action recognition algorithms trained, validated and tested on a subset of human behavior video sequences from Bosch's internal datasets. The anomaly detection network architecture defines how to properly reconstruct normal frame sequences so that during testing, each sequence can be classified as normal or abnormal based on its reconstruction error. With these errors, regularity scores are inferred showing the predicted regularity of each frame. The resulting framework is a viable addition to traditional action recognition algorithms since it can work as a tool for detecting unknown actions, strange/violent behaviors and aid in understanding the meaning of such human interactions.

Despite recent efforts, accuracy in group emotion recognition is still generally low. One of the reasons for these underwhelming performance levels is the scarcity of available labeled data which, like the literature approaches, is mainly focused on still images. In this work, we address this problem by adapting an inflated ResNet-50 pretrained for a similar task, activity recognition, where large labeled video datasets are available. Audio information is processed using a Bidirectional Long Short-Term Memory (Bi-LSTM) network receiving extracted features. A multimodal approach fuses audio and video information at the score level using a support vector machine classifier. Evaluation with data from the EmotiW 2020 AV Group-Level Emotion sub-challenge shows a final test accuracy of 65.74% for the multimodal approach, approximately 18% higher than the official baseline. The results show that using activity recognition pretraining offers performance advantages for group-emotion recognition and that audio is essential to improve the accuracy and robustness of video-based recognition.

Falls are among the frequent causes of the loss of mobility and independence in the elderly population. Given the global population aging, new strategies for predicting falls are required to reduce the number of their occurrences. In this study, a multifactorial screening protocol was applied to 281 community-dwelling adults aged over 65, and their 12-month prospective falls were annotated. Clinical and self-reported data, along with data from instrumented functional tests, involving inertial sensors and a pressure platform, were fused using early, late, and slow fusion approaches. For the early and late fusion, a classification pipeline was designed employing stratified sampling for the generation of the training and test sets. Grid search with cross-validation was used to optimize a set of feature selectors and classifiers. According to the slow fusion approach, each data source was mixed in the middle layers of a multilayer perceptron. The three studied fusion approaches yielded similar results for the majority of the metrics. However, if recall is considered to be more important than specificity, then the result of the late fusion approach providing a recall of [Formula: see text] is better compared with the results achieved by the other two approaches.

Sign Language Recognition (SLR) has become an appealing topic in modern societies because such technology can ideally be used to bridge the gap between deaf and hearing people. Although important steps have been made towards the development of real-world SLR systems, signer-independent SLR is still one of the bottleneck problems of this research field. In this regard, we propose a deep neural network along with an adversarial training objective, specifically designed to address the signer-independent problem. Concretely speaking, the proposed model consists of an encoder, mapping from input images to latent representations, and two classifiers operating on these underlying representations: (i) the signclassifier, for predicting the class/sign labels, and (ii) the signer-classifier, for predicting their signer identities. During the learning stage, the encoder is simultaneously trained to help the sign-classifier as much as possible while trying to fool the signer-classifier. This adversarial training procedure allows learning signer-invariant latent representations that are in fact highly discriminative for sign recognition. Experimental results demonstrate the effectiveness of the proposed model and its capability of dealing with the large inter-signer variations.

We propose a framework to model the distribution of sequential data coming from a set of entities connected in a graph with a known topology. The method is based on a mixture of shared hidden Markov models (HMMs), which are jointly trained in order to exploit the knowledge of the graph structure and in such a way that the obtained mixtures tend to be sparse. Experiments in different application domains demonstrate the effectiveness and versatility of the method.

In order to increase the adoption of machine learning models in areas like medicine and finance, it is necessary to have correct and diverse explanations for the decisions that the models provide, to satisfy the curiosity of decision-makers and the needs of the regulators. In this paper, we introduced a method, based in a previously presented framework, to explain the decisions of an Ensemble Model. Moreover, we instantiate the proposed approach to an ensemble composed of a Scorecard, a Random Forest, and a Deep Neural Network, to produce accurate decisions along with correct and diverse explanations. Our methods are tested on two biomedical datasets and one financial dataset. The proposed ensemble leads to an improvement in the quality of the decisions, and in the correctness of the explanations, when compared to its constituents alone. Qualitatively, it produces diverse explanations that make sense and convince the experts.

Active contour models are one of the most emblematic algorithms of computer vision. Their strong theoretical foundations and high user interoperability turned them into a reference approach for object segmentation and tracking tasks. A high number of modifications have already been proposed in order to overcome the known problems of traditional snakes, such as initialization dependence and poor convergence to concavities. In this paper, we address the scenario where the user wants to segment an object that has multiple dynamic regions but some of them do not correspond to the true object boundary. We propose a novel parametric active contour model, the Sparse Multi-Bending snake, which is capable of dividing the contour into a set of contiguous regions with different bending properties. We derive a new energy function that induces such behavior and presents a group optimization strategy that can be used to find the optimal bending resistance parameter for each point of the contour. We show the flexibility of our model in a set of synthetic images. In addition, we consider two real applications, lung segmentation in Computerized Tomography data and hand segmentation in depth images. We show how the proposed method is able to improve the segmentations obtained in both applications, when compared with other active contour models.

Ordinal arrangement of objects is a common property in biomedical images. Traditional methods to deal with semantic image segmentation in this setting are ad-hoc and application specific. In this paper, we propose ordinal-aware deep learning architectures for image segmentation that enforce pixelwise consistency by construction. We validated the proposed architectures on several real-life biomedical datasets and achieved competitive results in all cases.

Traditionally, convolutional neural networks are trained for semantic segmentation by having an image given as input and the segmented mask as output. In this work, we propose a neural network trained by being given an image and mask pair, with the output being the quality of that pairing. The segmentation is then created afterwards through backpropagation on the mask. This allows enriching training with semi-supervised synthetic variations on the ground-truth. The proposed iterative segmentation technique allows improving an existing segmentation or creating one from scratch. We compare the performance of the proposed methodology with state-of-the-art deep architectures for image segmentation and achieve competitive results, being able to improve their segmentations.

Ordinal classification is a specific and demanding task, where the aim is not only to increase accuracy, but to also capture the natural order between the classes, and penalize incorrect predictions by how much they deviate from this ranking. If an ordinal classifier must be able to comply with all these requirements, a suitable ordinal metric must be able to accurately measure its degree of compliance. However, the current metrics are unable to completely capture these considerations when assessing classification performance. Moreover, most suffer from sensitivity to imbalanced classes, very common in ordinal classification. In this paper, we propose two variants of a novel performance index that accounts for both accuracy and ranking in the performance assessment of ordinal classification, and is robust against imbalanced classes.

Through the years, several CAD systems have been developed to help radiologists in the hard task of detecting signs of cancer in mammograms. In these CAD systems, mass segmentation plays a central role in the decision process. In the literature, mass segmentation has been typically evaluated in a intra-sensor scenario, where the methodology is designed and evaluated in similar data. However, in practice, acquisition systems and PACS from multiple vendors abound and current works fails to take into account the differences in mammogram data in the performance evaluation. In this work it is argued that a comprehensive assessment of the mass segmentation methods requires the design and evaluation in datasets with different properties. To provide a more realistic evaluation, this work proposes: a) improvements to a state of the art method based on tailored features and a graph model; b) a head-to-head comparison of the improved model with recently proposed methodologies based in deep learning and structured prediction on four reference databases, performing a cross-sensor evaluation. The results obtained support the assertion that the evaluation methods from the literature are optimistically biased when evaluated on data gathered from exactly the same sensor and/or acquisition protocol.

In classification, when there is a disproportion in the number of observations in each class, the data is said to be class imbalance. Class imbalance is pervasive in real world applications of data classification and has been the focus of much research. The minority class contributes too little to the decision boundary because the learning process learns from each observation in isolation. In this paper, we discuss the application of learning pairwise rankers as a solution to class imbalance. We compare ranking models to alternatives from the literature.

We study the problem of learning lexicographic preferences on multiattribute domains, and propose Rankdom Forests as a compact way to express preferences in learning to rank scenarios. We start generalizing Conditional Lexicographic Preference Trees by introducing multiple kernels in order to handle non-categorical attributes. Then, we define a learning strategy for inferring lexicographic rankers from partial pairwise comparisons between options. Finally, a Lexicographic Ensemble is introduced to handle multiple weak partial rankers, being Rankdom Forests one of these ensembles. We tested the performance of the proposed method using several datasets and obtained competitive results when compared with other lexicographic rankers.

Generalized additive models are well-known as a powerful and palatable predictive modelling technique. Scorecards, the discretized version of generalized additive models, are a long-established method in the industry, due to its balance between simplicity and performance. Scorecards are easy to app ly and easy to understand. Moreover, in spite of their simplicity, scorecards can model nonlinear relationships between the inputs and the value to be predicted. In the scientific community, scorecards have been largely overlooked in favor of more recent models such as neural networks or support vector machines. In this paper, we address scorecard development, introducing a new formulation more suitable to support regularization. We tackle both the binary and the ordinal data classification problems. In both settings, the proposed methodology shows advantages when evaluated using real datasets.

Breast cancer is one of the most mediated malignant diseases, because of its high incidence and prevalence, but principally due to its physical and psychological invasiveness. The study of this disease using computer science tools resorts often to the image segmentation operation. Image segmentation, although having been extensively studied, is still an open problem. Shortest path algorithms are extensively used to tackle this problem. There are, however, applications where the starting and ending positions of the shortest path need to be constrained, defining a closed contour enclosing a previously detected seed. Mass and calcification segmentation in mammograms and areola segmentation in digital images are two particular examples of interest within the field of breast cancer research. Usually the closed contour computation is addressed by transforming the image into polar coordinates, where the closed contour is transformed into an open contour between two opposite margins. In this work, after illustrating some of the limitations of this approach, we show how to compute the closed contour in the original coordinate space. After defining a directed acyclic graph appropriate for this task, we address the main difficulty in operating in the original coordinate space. Since small paths collapsing in the seed point are naturally favored, we modulate the cost of the edges to counterbalance this bias. A thorough evaluation is conducted with datasets from the breast cancer field. The algorithm is shown to be fast and reliable and suffers no loss in resolution.

Through the years, several CAD systems have been developed to help radiologists in the hard task of detecting signs of cancer in the numerous screening mammograms. A more recent trend includes the development of pre-CAD systems aiming at identifying normal mammograms instead of detecting suspicious ones. Normal breasts are screened-out from the process, leaving radiologists more time to focus on more difficult cases. In this work, a new approach for the identification of normal breasts is presented. Considering that even breasts with malignant findings are mostly constituted by normal tissue, the breast area is divided into blocks which are then compared pairwise. If all blocks are very similar, the breast is labelled as normal, and as suspicious otherwise. Features characterizing the pairwise block similarity and characterizing the intra-block pixel distribution are used to design a predictive method based on machine learning techniques. The proposed solution was applied on a real world screening setting composed by nearly 18000 mammograms. Results are similar to the more complex state of the art approaches by correctly identifying more than 20% of the normal mammograms. These results suggest the usefulness of the relative comparison instead of the absolute classification. When properly used, simple statistics can suffice to distinguish the clearly normal breasts. © 2014 IEEE.

Breast cancer is one of the most mediated malignant diseases, because of its high incidence and prevalence, but principally due to its physical and psychological invasiveness. Surgeons and patients have often many options to consider for undergoing the procedure. The ability to visualise the potential outcomes of the surgery and make decisions on their surgical options is, therefore, very important for patients and surgeons. In this paper we investigate the fitting of a 3d point cloud of the breast to a parametric model usable in surgery planning, obtaining very promising results with real data.

Biometric systems based on iris are vulnerable to several attacks, particularly direct attacks consisting on the presentation of a fake iris to the sensor. The development of iris liveness detection techniques is crucial for the deployment of iris biometric applications in daily life specially in the mobile biometric field. The 1stMobile Iris Liveness Detection Competition (MobILive) was organized in the context of IJCB2014 in order to record recent advances in iris liveness detection. The goal for (MobILive) was to contribute to the state of the art of this particular subject. This competition covered the most common and simple spoofing attack in which printed images from an authorized user are presented to the sensor by a non-authorized user in order to obtain access. The benchmark dataset was the MobBIOfake database which is composed by a set of 800 iris images and its corresponding fake copies (obtained from printed images of the original ones captured with the same handheld device and in similar conditions). In this paper we present a brief description of the methods and the results achieved by the six participants in the competition.

Animal behavior assessment plays an important role in basic and clinical neuroscience. Although assessing the higher functional level of the nervous system is already possible, behavioral tests are extremely complex to design and analyze. Animal's responses are often evaluated manually, making it subjective, extremely time consuming, poorly reproducible and potentially fallible. The main goal of the present work is to evaluate the use of consumer depth cameras, such as the Microsoft's Kinect, for detection of behavioral patterns of mice. The hypothesis is that the depth information, should enable a more feasible and robust method for automatic behavior recognition. Thus, we introduce our depth-map based approach comprising mouse segmentation, body-like per-frame feature extraction and per-frame classification given temporal context, to prove the usability of this methodology.

While video surveillance systems are spreading everywhere, extracting meaningful information from what they are recording is still prohibitively expensive. There is a major effort under way in order to make this process economical by including an intelligent software that eases the burden of the system. In this paper, we introduce an incremental learning framework to classify parallel data streams generated in a multi-camera surveillance scenario. The framework exploits active learning strategies in order to interact wisely with operators to address various problems that exist in such non-stationary environments, such as concept drift and concept evolution. If we look at the problem as mining parallel streams, the framework can address learning from uneven parallel streams applying a class-based ensemble, a problem that has not been addressed before. Favourable results indicate the success of the framework.

Biometrie systems based on iris are vulnerable to direct attacks consisting on the presentation of a fake iris to the sensor (a printed or a contact lenses iris image, among others). The mobile biometrics scenario stresses the importance of assessing the security issues. The application of countermeasures against this type of attacking scheme is the problem addressed in the present paper. Widening a previous work, several state-of-the-art iris liveness detection methods were implemented and adapted to a less-constrained scenario. The proposed method combines a feature selection step prior to the use of state-of-the-art classifiers to perform the classification based upon the "best features". Five well known existing databases for iris liveness purposes (Biosec, Clarkson, NotreDame and Warsaw) and a recently published database, MobBIOfake, with real and fake images captured in the mobile scenario were tested. The results obtained suggest that the automated segmentation step does not degrade significantly the results.

The increasing demand for human activity analysis on surveillance scenarios has been provoking the emerging of new features and concepts that could help to identify the activities of interest. In this paper, we present a context-based descriptor to identify individual profiles. It accounts with a multi-scale histogram representation of position-based and attention-based features that follow a key-point trajectory sampling. The notion of profile is expressed by a new semantic concept introduced as an adjective for action recognition. We also identify a very rich dataset, in terms of intensity and variability of human activity, and extended it by manual annotation to validate the introduced concept of profile and test the descriptor's discriminative power. High rates of recognition were achieved.

In predictive modeling tasks, knowledge about the training examples is neither fully complete nor totally incomplete. Unlike semisupervised learning, where one either has perfect knowledge about the label of the point or is completely ignorant about it, here we address a setting where, for each example, we only possess partial information about the label. Each example is described using two (or more) different feature sets or views, where neither are necessarily observed for a given example. If a single view is observed, then the class is only due to that feature set; if more views are present, the observed class label is the maximum of the values corresponding to the individual views. After formalizing this new learning concept, we propose two new learning methodologies that are adapted to this learning paradigm. We also compare their instantiation in experiments with different base models and with conventional methods. The experimental results made both on real and synthetic data sets verify the usefulness of the proposed approaches.

The detection of staff lines is the first step of most Optical Music Recognition (OMR) systems. Its great significance derives from the ease with which we can then proceed with the extraction of musical symbols. All OMR tasks are usually achieved using binary images by setting thresholds that can be local or global. These techniques however, may remove relevant information of the music sheet and introduce artifacts which will degrade results in the later stages of the process. It arises therefore a need to create a method that reduces the loss of information due to the binarization. The baseline for the methodology proposed in this paper follows the shortest path algorithm proposed in [CardosoTPAMI08]. The concept of strong staff pixels (SSP's), which is a set of pixels with a high probability of belonging to a staff line, is proposed to guide the cost function. The SSP allows to overcome the results of the binary based detection and to generalize the binary framework to grayscale music scores. The proposed methodology achieves good results.

Breast Cancer Conservative Treatment (BCCT) is now the preferred technique for breast cancer treatment. The limited reproducibility of standard aesthetic evaluation methods led to the development of objective methods, such as Breast Cancer Conservative Treatment.cosmetic results (BCCT.core) software tool. Although the satisfying results, there are still limitations concerning complete automation and the inability to measure volumetric information. With the fundamental premise of maintaining the system as a low-cost tool, the incorporation of the Microsoft Kinect sensor in BCCT evaluations was studied. The aim with this work is to enable the simultaneous detection of breast contour and breast peak points using depth-map data. Experimental results show that the proposed algorithm is accurate and robust for a wide number of patients. Additionally, comparatively to previous research, the procedure for detecting prominent points was automated.

In this paper we present a complete system for object tracking over multiple uncalibrated cameras with or without overlapping fields of view. We employ an approach based on the bag-of-visterms technique to represent and match tracked objects. The tracks are compared with a global object model based on an ensemble of individual object models. The system can globally recognise objects and minimise common tracking problems such as track drift or split. The output is a timeline representing the objects present in a given multi-camera scene. The methods employed in the system are online and can be optimized to operate in real-time.

Ordinal data classification (ODC) has a wide range of applications in areas where human evaluation plays an important role, ranging from psychology and medicine to information retrieval. In ODC the output variable has a natural order; however, there is not a precise notion of the distance between classes. The recently proposed method for ordinal data, Kernel Discriminant Learning Ordinal Regression (KDLOR), is based on Linear Discriminant Analysis (LDA), a simple tool for classification. KDLOR brings LDA to the forefront in the ODC held, motivating further research. This paper compares three LDA based algorithms for ODC. The first method uses the generic framework of Frank and Hall for ODC instantiated with a kernel version of LDA. Similarly, the second method is based on the also generic Data Replication framework for ODC instantiated with the same kernel version of LDA. Both the Frank and Hall and Data Replication methods address the ODC problem by the use of a base binary classifier. Finally, the third method under comparison is KDLOR. The experiments are carried out on synthetic and real datasets. A comparison between the performances of the three systems is made based on tstatistics. The performance and running time complexity of the methods do not support any advantage of KDLOR over the other two methods.

While ordinal classification problems are common in many situations, induction of ordinal decision trees has not evolved significantly. Conventional trees for regression settings or nominal classification are commonly induced for ordinal classification problems. On the other hand a decision tree consistent with the ordinal setting is often desirable to aid decision making in such situations as credit rating. In this work we extend a recently proposed strategy based on constraints defined globally over the feature space. We propose a bootstrap technique to improve the accuracy of the baseline solution. Experiments in synthetic and real data show the benefits of our proposal.

Ordinal classification is a form of multiclass classification for which there is an inherent order between the classes, but not a meaningful numeric difference between them. The performance of such classifiers is usually assessed by measures appropriate for nominal classes or for regression. Unfortunately, these do not account for the true dimension of the error. The goal of this work is to show that existing measures for evaluating ordinal classification models suffer from a number of important shortcomings. For this reason, we propose an alternative measure defined directly in the confusion matrix. An error coefficient appropriate for ordinal data should capture how much the result diverges from the ideal prediction and how "inconsistent" the classifier is in regard to the relative order of the classes. The proposed coefficient results from the observation that the performance yielded by the Misclassification Error Rate coefficient is the benefit of the path along the diagonal of the confusion matrix. We carry out an experimental study which confirms the usefulness of the novel metric.

Ordinal classification is a form of multi-class classification where there is an inherent ordering between the classes, but not a meaningful numeric difference between them. Although conventional methods, designed for nominal classes or regression problems, can be used to solve the ordinal data problem, there are benefits in developing models specific to this kind of data. This paper introduces a new rationale to include the information about the order in the design of a classification model. The method encompasses the inclusion of consistency constraints between adjacent decision regions. A new decision tree and a new nearest neighbour algorithms are then designed under that rationale. An experimental study with artificial and real data sets verifies the usefulness of the proposed approach.

Support vector machines (SVMs) were initially proposed to solve problems with two classes. Despite the myriad of schemes for multiclassification with SVMs proposed since then, little work has been done for the case where the classes are ordered. Usually one constructs a nominal classifier and a posteriori defines the order. The definition of an ordinal classifier leads to a better generalisation. Moreover, most of the techniques presented so far in the literature can generate ambiguous regions. All-at-Once methods have been proposed to solve this issue. In this work we devise a new SVM methodology based on the unimodal paradigm with the All-at-Once scheme for the ordinal classification.

The optical recognition of handwritten musical scores by computers remains far from ideal. Most OMR algorithms rely on an estimation of the staff line thickness and the vertical line distance within the same staff. Subsequent operation can use these values as references, dismissing the need for some predetermined threshold values. In this work we improve on previous conventional estimates for these two reference lengths. We start by proposing a new method for binarized music scores and then extend the approach for gray-level music scores. An experimental study with 50 images is used to assess the interest of the novel method.

In this work we consider the problem of binary classification where the classifier may abstain instead of classifying each observation, leaving the critical items for human evaluation. This article motivates and presents a novel method to learn the reject region on complex data. Observations are replicated and then a single binary classifier determines the decision plane. The proposed method is an extension of a method available in the literature for the classification of ordinal data. Our method is compared with standard techniques on synthetic and real datasets, emphasizing the advantages of the proposed approach.

Text line segmentation in freestyle handwritten documents remains an open document analysis problem. Curvilinear text lines and small gaps between neighbouring text lines present a challenge to algorithms developed for machine-printed or hand-printed documents. We investigate a general-purpose, knowledge-free method for the automatic detection of text lines based on a stable path approach. Lines affected by curvature and inclination are robustly detected. The proposed methodology was tested on a modern set of handwritten images made available on the ICDAR 2009 handwriting segmentation competition, with promissing results.

The preservation of many music works produced in the past entails their digitalization and consequent accessibility in an easy-to-manage digital format. Carrying this task manually is very time consuming and error prone. While optical music recognition systems usually perform well on printed scores, the processing of handwritten musical scores by computers remain far from ideal. One of the fundamental stages to carry out this task is the staff line detection. In this paper a new method for the automatic detection of music staff lines based on a connected path approach is presented. Lines affected by curvature, discontinuities, and inclination are robustly detected. Experimental results show that the proposed technique consistently outperforms well-established algorithms.

Breast cancer conservative treatment (BCCT) is considered the gold standard of breast cancer treatment. However, aesthetic results are heterogeneous and difficult to evaluate in a standardised way. The limited reproducibility of subjective aesthetic evaluation in BCCT forced the research on objective methods. A recent computer system was developed to objectively and automatically evaluate the aesthetic result of BCCT. In this system, the detection of the breast contour on the digital photograph of the patient is necessary to extract the features subsequently used in the evaluation process. In this paper we extend an algorithm based on the shortest path on a graph to detect automatically the breast contour. The advantage of graph algorithms is that they are guaranteed to find the global optimum of the problem; the difficulty is that they make it hard to enforce shape constraints. We define and compare different techniques to introduce the a priory knowledge of the mammary contour. Experimental results show that the proposed techniques consistently outperform the base method.

Byte stuffing is a technique to allow the transparent transmission of arbitrary sequences with constrained sequences. To date, most of the existing algorithms, such as PPP, attain a low average overhead by sacrificing the worst-case scenario. An exception is COBS which was designed for a low worst-case overhead; however, it imposes always a nonzero overhead, even on small packets. In this work is proposed a byte stuffing algorithm that simultaneously controls the average and worst-case overhead, performing close to the theoretical bound. It is shown analytically that the proposed algorithm achieves improved average and worst-case rates over state of the art methods. Furthermore, this technique is generalized to hybrid methods, with lower computing complexity. It is further analysed and compared experimentally the behaviour of the proposed algorithm against established algorithms in terms of byte overhead and computational time.

This paper addresses the problem of unsupervised speaker change detection. Three systems based on the Bayesian Information Criterion (BIC) are tested. The first system investigates the AudioSpectrumCentroid and the AudioWaveformEnvelope features, implements a dynamic thresholding followed by a fusion scheme, and finally applies BIC. The second method is a real-time one that uses a metric-based approach employing the line spectral pairs and the BIC to validate a potential speaker change point. The third method consists of three modules. In the first module, a measure based on second-order statistics is used; in the second module, the Euclidean distance and T2 Hotelling statistic are applied; and in the third module, the BIC is utilized. The experiments are carried out on a dataset created by concatenating speakers from the TIMIT database, that is referred to as the TIMIT data set. A comparison between the performance of the three systems is made based on t-statistics.

In this paper, we recover a graph interpretation of the mutual partition distance, proposed recently by Cardoso and Corte-Real. We deduce some properties of this measure, and establish a correspondence with the partition distance introduced by Almudevar and Field and Gusfield, and independently by Guigues. We also present some different formulations for the computation of the mutual partition distance. Finally, a comparison is made with similar measures.

Cosmetic assessment of conservative breast cancer treatment plays a major role in the study of breast cancer techniques. Objective assessment methods are being preferred to overcome the drawbacks of subjective evaluation. In this paper a methodology for the objective assessment of conservative breast cancer treatment is proposed. The quantitative measures used in this research provide an objective way to calculate the overall cosmetic result. We report experiments using support vector machines to derive an optimal assessment rule. The results seem to indicate that it is possible to construct an algorithm for a complete objective classification of the aesthetic result of breast conservative treatment.

The implementation of fast dedicated processor for block matching motion estimation based on cumulants matching criteria implies the optimization of all of its components. Special care should be spent with the multiply-accumulate unit that is the core of many digital signal processing systems. Therefore, its optimization may be of outmost importance, specially if a significative number of such units are present in the platform. In this paper, the minimization of the size of one such unit is provided for a specific application, although the results have relevance in other scenarios.

Image segmentation plays a major role in a broad range of applications. Evaluating the adequacy of a segmentation algorithm for a given application is a requisite both to allow the appropriate selection of segmentation algorithms as well as to tune their parameters for optimal performance. However, objective segmentation quality evaluation is far from being a solved problem. In this paper, a generic framework for segmentation evaluation is introduced after a brief review of previous work. A metric based on the distance between segmentation partitions is proposed to overcome some of the limitations of existing approaches. Symmetric and asymmetric distance metric alternatives are presented to meet the specificities of a wide class of applications. Experimental results confirm the potential of the proposed measures.