arXiv subjects

Deniz Erdogmus

Publications and source records attributed to Deniz Erdogmus.

70 records · Page 4Linked to original sources

arXiv2018

Asymmetric Loss Functions and Deep Densely Connected Networks for Highly Imbalanced Medical Image Segmentation: Application to Multiple Sclerosis Lesion Detection

Fully convolutional deep neural networks have been asserted to be fast and precise frameworks with great potential in image segmentation. One of the major challenges in training such networks raises when data is unbalanced, which is common in many medical imaging applications such as lesion segmentation where lesion class voxels are often much lower in numbers than non-lesion voxels. A trained network with unbalanced data may make predictions with high precision and low recall, being severely biased towards the non-lesion class which is particularly undesired in most medical applications where FNs are more important than FPs. Various methods have been proposed to address this problem, more recently similarity loss functions and focal loss. In this work we trained fully convolutional deep neural networks using an asymmetric similarity loss function to mitigate the issue of data imbalance and achieve much better tradeoff between precision and recall. To this end, we developed a 3D FC-DenseNet with large overlapping image patches as input and an asymmetric similarity loss layer based on Tversky index (using Fbeta scores). We used large overlapping image patches as inputs for intrinsic and extrinsic data augmentation, a patch selection algorithm, and a patch prediction fusion strategy using B-spline weighted soft voting to account for the uncertainty of prediction in patch borders. We applied this method to MS lesion segmentation based on two different datasets of MSSEG and ISBI longitudinal MS lesion segmentation challenge, where we achieved top performance in both challenges. Our network trained with focal loss ranked first according to the ISBI challenge overall score and resulted in the lowest reported lesion false positive rate among all submitted methods. Our network trained with the asymmetric similarity loss led to the lowest surface distance and the best lesion true positive rate.

cs.CV↗

arXiv2018

Hierarchical Graphical Models for Context-Aware Hybrid Brain-Machine Interfaces

We present a novel hierarchical graphical model based context-aware hybrid brain-machine interface (hBMI) using probabilistic fusion of electroencephalographic (EEG) and electromyographic (EMG) activities. Based on experimental data collected during stationary executions and subsequent imageries of five different hand gestures with both limbs, we demonstrate feasibility of the proposed hBMI system through within session and online across sessions classification analyses. Furthermore, we investigate the context-aware extent of the model by a simulated probabilistic approach and highlight potential implications of our work in the field of neurophysiologically-driven robotic hand prosthetics.

cs.HC↗

arXiv2018

Time-Series Prediction of Proximal Aggression Onset in Minimally-Verbal Youth with Autism Spectrum Disorder Using Physiological Biosignals

It has been suggested that changes in physiological arousal precede potentially dangerous aggressive behavior in youth with autism spectrum disorder (ASD) who are minimally verbal (MV-ASD). The current work tests this hypothesis through time-series analyses on biosignals acquired prior to proximal aggression onset. We implement ridge-regularized logistic regression models on physiological biosensor data wirelessly recorded from 15 MV-ASD youth over 64 independent naturalistic observations in a hospital inpatient unit. Our results demonstrate proof-of-concept, feasibility, and incipient validity predicting aggression onset 1 minute before it occurs using global, person-dependent, and hybrid classifier models.

cs.HC↗

arXiv2018

Real-time Deep Pose Estimation with Geodesic Loss for Image-to-Template Rigid Registration

With an aim to increase the capture range and accelerate the performance of state-of-the-art inter-subject and subject-to-template 3D registration, we propose deep learning-based methods that are trained to find the 3D position of arbitrarily oriented subjects or anatomy based on slices or volumes of medical images. For this, we propose regression CNNs that learn to predict the angle-axis representation of 3D rotations and translations using image features. We use and compare mean square error and geodesic loss to train regression CNNs for 3D pose estimation used in two different scenarios: slice-to-volume registration and volume-to-volume registration. Our results show that in such registration applications that are amendable to learning, the proposed deep learning methods with geodesic loss minimization can achieve accurate results with a wide capture range in real-time (<100ms). We also tested the generalization capability of the trained CNNs on an expanded age range and on images of newborn subjects with similar and different MR image contrasts. We trained our models on T2-weighted fetal brain MRI scans and used them to predict the 3D pose of newborn brains based on T1-weighted MRI scans. We showed that the trained models generalized well for the new domain when we performed image contrast transfer through a conditional generative adversarial network. This indicates that the domain of application of the trained deep regression CNNs can be further expanded to image modalities and contrasts other than those used in training. A combination of our proposed methods with accelerated optimization-based registration algorithms can dramatically enhance the performance of automatic imaging devices and image processing methods of the future.

cs.CV↗

arXiv2018

Invariant Representations from Adversarially Censored Autoencoders

We combine conditional variational autoencoders (VAE) with adversarial censoring in order to learn invariant representations that are disentangled from nuisance/sensitive variations. In this method, an adversarial network attempts to recover the nuisance variable from the representation, which the VAE is trained to prevent. Conditioning the decoder on the nuisance variable enables clean separation of the representation, since they are recombined for model learning and data reconstruction. We show this natural approach is theoretically well-founded with information-theoretic arguments. Experiments demonstrate that this method achieves invariance while preserving model learning performance, and results in visually improved performance for style transfer and generative sampling tasks.

cs.LG↗

arXiv2017

EEG-assisted Modulation of Sound Sources in the Auditory Scene

Noninvasive EEG (electroencephalography) based auditory attention detection could be useful for improved hearing aids in the future. This work is a novel attempt to investigate the feasibility of online modulation of sound sources by probabilistic detection of auditory attention, using a noninvasive EEG-based brain computer interface. Proposed online system modulates the upcoming sound sources through gain adaptation which employs probabilistic decisions (soft decisions) from a classifier trained on offline calibration data. In this work, calibration EEG data were collected in sessions where the participants listened to two sound sources (one attended and one unattended). Cross-correlation coefficients between the EEG measurements and the attended and unattended sound source envelope (estimates) are used to show differences in sharpness and delays of neural responses for attended versus unattended sound source. Salient features to distinguish attended sources from the unattended ones in the correlation patterns have been identified, and later they have been used to train an auditory attention classifier. Compared to the existing results in the literature, in this paper we have two main contributions. First, using the auditory attention classifier, we have shown high offline detection performance with single channel EEG measurements of shorter duration compared to the existing approaches in the literature which employ large number of channels with longer EEG measurements. Second, using the classifier trained offline in the calibration session, we have shown the performance of the online sound source modulation system. We observe that online sound source modulation system is able to keep the level of attended sound source higher than the unattended source. Keywords: auditory BCI, cocktail party problem, auditory attention classification

q-bio.NC↗

arXiv2017

Real-Time Automatic Fetal Brain Extraction in Fetal MRI by Deep Learning

Brain segmentation is a fundamental first step in neuroimage analysis. In the case of fetal MRI, it is particularly challenging and important due to the arbitrary orientation of the fetus, organs that surround the fetal head, and intermittent fetal motion. Several promising methods have been proposed but are limited in their performance in challenging cases and in real-time segmentation. We aimed to develop a fully automatic segmentation method that independently segments sections of the fetal brain in 2D fetal MRI slices in real-time. To this end, we developed and evaluated a deep fully convolutional neural network based on 2D U-net and autocontext, and compared it to two alternative fast methods based on 1) a voxelwise fully convolutional network and 2) a method based on SIFT features, random forest and conditional random field. We trained the networks with manual brain masks on 250 stacks of training images, and tested on 17 stacks of normal fetal brain images as well as 18 stacks of extremely challenging cases based on extreme motion, noise, and severely abnormal brain shape. Experimental results show that our U-net approach outperformed the other methods and achieved average Dice metrics of 96.52% and 78.83% in the normal and challenging test sets, respectively. With an unprecedented performance and a test run time of about 1 second, our network can be used to segment the fetal brain in real-time while fetal MRI slices are being acquired. This can enable real-time motion tracking, motion detection, and 3D reconstruction of fetal brain MRI.

cs.CV↗

arXiv2017

Auto-context Convolutional Neural Network (Auto-Net) for Brain Extraction in Magnetic Resonance Imaging

Brain extraction or whole brain segmentation is an important first step in many of the neuroimage analysis pipelines. The accuracy and robustness of brain extraction, therefore, is crucial for the accuracy of the entire brain analysis process. With the aim of designing a learning-based, geometry-independent and registration-free brain extraction tool in this study, we present a technique based on an auto-context convolutional neural network (CNN), in which intrinsic local and global image features are learned through 2D patches of different window sizes. In this architecture three parallel 2D convolutional pathways for three different directions (axial, coronal, and sagittal) implicitly learn 3D image information without the need for computationally expensive 3D convolutions. Posterior probability maps generated by the network are used iteratively as context information along with the original image patches to learn the local shape and connectedness of the brain, to extract it from non-brain tissue. The brain extraction results we have obtained from our algorithm are superior to the recently reported results in the literature on two publicly available benchmark datasets, namely LPBA40 and OASIS, in which we obtained Dice overlap coefficients of 97.42% and 95.40%, respectively. Furthermore, we evaluated the performance of our algorithm in the challenging problem of extracting arbitrarily-oriented fetal brains in reconstructed fetal brain magnetic resonance imaging (MRI) datasets. In this application our algorithm performed much better than the other methods (Dice coefficient: 95.98%), where the other methods performed poorly due to the non-standard orientation and geometry of the fetal brain in MRI. Our CNN-based method can provide accurate, geometry-independent brain extraction in challenging applications.

cs.CV↗

arXiv2017

Tversky loss function for image segmentation using 3D fully convolutional deep networks

Fully convolutional deep neural networks carry out excellent potential for fast and accurate image segmentation. One of the main challenges in training these networks is data imbalance, which is particularly problematic in medical imaging applications such as lesion segmentation where the number of lesion voxels is often much lower than the number of non-lesion voxels. Training with unbalanced data can lead to predictions that are severely biased towards high precision but low recall (sensitivity), which is undesired especially in medical applications where false negatives are much less tolerable than false positives. Several methods have been proposed to deal with this problem including balanced sampling, two step training, sample re-weighting, and similarity loss functions. In this paper, we propose a generalized loss function based on the Tversky index to address the issue of data imbalance and achieve much better trade-off between precision and recall in training 3D fully convolutional deep neural networks. Experimental results in multiple sclerosis lesion segmentation on magnetic resonance images show improved F2 score, Dice coefficient, and the area under the precision-recall curve in test data. Based on these results we suggest Tversky loss function as a generalized framework to effectively train deep neural networks.

cs.CV↗

arXiv2017

Subject-Specific Abnormal Region Detection in Traumatic Brain Injury Using Sparse Model Selection on High Dimensional Diffusion Data

We present a method to estimate a multivariate Gaussian distribution of diffusion tensor features in a set of brain regions based on a small sample of healthy individuals, and use this distribution to identify imaging abnormalities in subjects with mild traumatic brain injury. The multivariate model receives a {\em apriori} knowledge in the form of a neighborhood graph imposed on the precision matrix, which models brain region interactions, and an additional $L_1$ sparsity constraint. The model is then estimated using the graphical LASSO algorithm and the Mahalanobis distance of healthy and TBI subjects to the distribution mean is used to evaluate the discriminatory power of the model. Our experiments show that the addition of the {\em apriori} neighborhood graph results in significant improvements in classification performance compared to a model which does not take into account the brain region interactions or one which uses a fully connected prior graph. In addition, we describe a method, using our model, to detect the regions that contribute the most to the overall abnormality of the DTI profile of a subject's brain.

stat.AP↗

arXiv2017

Decoding Complex Imagery Hand Gestures

Brain computer interfaces (BCIs) offer individuals suffering from major disabilities an alternative method to interact with their environment. Sensorimotor rhythm (SMRs) based BCIs can successfully perform control tasks; however, the traditional SMR paradigms intuitively disconnect the control and real task, making them non-ideal for complex control scenarios. In this study, we design a new, intuitively connected motor imagery (MI) paradigm using hierarchical common spatial patterns (HCSP) and context information to effectively predict intended hand grasps from electroencephalogram (EEG) data. Experiments with 5 participants yielded an aggregate classification accuracy--intended grasp prediction probability--of 64.5\% for 8 different hand gestures, more than 5 times the chance level.

cs.HC↗

arXiv2017

Context-Aware Recursive Bayesian Graph Traversal in BCIs

Noninvasive brain computer interfaces (BCI), and more specifically Electroencephalography (EEG) based systems for intent detection need to compensate for the low signal to noise ratio of EEG signals. In many applications, the temporal dependency information from consecutive decisions and contextual data can be used to provide a prior probability for the upcoming decision. In this study we proposed two probabilistic graphical models (PGMs), using context information and previously observed EEG evidences to estimate a probability distribution over the decision space in graph based decision-making mechanism. In this approach, user moves a pointer to the desired vertex in the graph in which each vertex represents an action. To select a vertex, a Select command, or a proposed probabilistic Selection criterion (PSC) can be used to automatically detect the user intended vertex. Performance of different PGMs and Selection criteria combinations are compared over a keyboard based on a graph layout. Based on the simulation results, probabilistic Selection criterion along with the probabilistic graphical model provides the highest performance boost for individuals with pour calibration performance and achieving the same performance for individuals with high calibration performance.

cs.HC↗

arXiv2016

Asymptotic Analysis of Objectives based on Fisher Information in Active Learning

Obtaining labels can be costly and time-consuming. Active learning allows a learning algorithm to intelligently query samples to be labeled for efficient learning. Fisher information ratio (FIR) has been used as an objective for selecting queries in active learning. However, little is known about the theory behind the use of FIR for active learning. There is a gap between the underlying theory and the motivation of its usage in practice. In this paper, we attempt to fill this gap and provide a rigorous framework for analyzing existing FIR-based active learning methods. In particular, we show that FIR can be asymptotically viewed as an upper bound of the expected variance of the log-likelihood ratio. Additionally, our analysis suggests a unifying framework that not only enables us to make theoretical comparisons among the existing querying methods based on FIR, but also allows us to give insight into the development of new active learning approaches based on this objective.

stat.ML↗

arXiv2016

An Active RBSE Framework to Generate Optimal Stimulus Sequences in a BCI for Spelling

A class of brain computer interfaces (BCIs) employs noninvasive recordings of electroencephalography (EEG) signals to enable users with severe speech and motor impairments to interact with their environment and social network. For example, EEG based BCIs for typing popularly utilize event related potentials (ERPs) for inference. Presentation paradigm design in current ERP-based letter by letter typing BCIs typically query the user with an arbitrary subset characters. However, the typing accuracy and also typing speed can potentially be enhanced with more informed subset selection and flash assignment. In this manuscript, we introduce the active recursive Bayesian state estimation (active-RBSE) framework for inference and sequence optimization. Prior to presentation in each iteration, rather than showing a subset of randomly selected characters, the developed framework optimally selects a subset based on a query function. Selected queries are made adaptively specialized for users during each intent detection. Through a simulation-based study, we assess the effect of active-RBSE on the performance of a language-model assisted typing BCI in terms of typing speed and accuracy. To provide a baseline for comparison, we also utilize standard presentation paradigms namely, row and column matrix presentation paradigm and also random rapid serial visual presentation paradigms. The results show that utilization of active-RBSE can enhance the online performance of the system, both in terms of typing accuracy and speed.

cs.HC↗

arXiv2016

Fast Switch Scanning Keyboards: Minimal Expected Query Decision Trees

Augmentative and Alternative Communication (AAC) systems allow people with disabilities to provide input to devices which empower them to more fully interact with their environment. Within AAC, switch scanning is a common paradigm for spelling where a set of characters is highlighted and the user is queried as to whether their target character is in the highlighted set. These queries are used to traverse a decision tree which successively prunes away characters until only a single one remains (the estimate). This work seeks a decision tree which requires the fewest expected queries per decision sequence (EQPD). In particular, we remove the constraint that the decision tree needs to be a row-item or group-row-item style tree and minimize EQPD. We pose the problem as a Huffman code with variable, integer cost and solve it with a mild extension of Golin's method in "A dynamic programming algorithm for constructing optimal prefix-free codes with unequal letter costs", IEEE Transactions on Information Theory (1998). Additionally, we model the user on the query level by their probability of detection and false alarm to derive their expected performance on the character level given some decision tree. We perform experiments which show that the min EQPD decision tree (Karp) may reduce selection times, especially for timed (single switch) switch scanning.

cs.HC↗

arXiv2016

Manifold unwrapping using density ridges

Research on manifold learning within a density ridge estimation framework has shown great potential in recent work for both estimation and de-noising of manifolds, building on the intuitive and well-defined notion of principal curves and surfaces. However, the problem of unwrapping or unfolding manifolds has received relatively little attention within the density ridge approach, despite being an integral part of manifold learning in general. This paper proposes two novel algorithms for unwrapping manifolds based on estimated principal curves and surfaces for one- and multi-dimensional manifolds respectively. The methods of unwrapping are founded in the realization that both principal curves and principal surfaces will have inherent local maxima of the probability density function. Following this observation, coordinate systems that follow the shape of the manifold can be computed by following the integral curves of the gradient flow of a kernel density estimate on the manifold. Furthermore, since integral curves of the gradient flow of a kernel density estimate is inherently local, we propose to stitch together local coordinate systems using parallel transport along the manifold. We provide numerical experiments on both real and synthetic data that illustrates clear and intuitive unwrapping results comparable to state-of-the-art manifold learning algorithms.

stat.ML↗