Prognoses of Traumatic Brain Injury (TBI) outcomes are neither easily nor accurately determined from clinical indicators. This is due in part to the heterogeneity of damage inflicted to the brain, ultimately resulting in diverse and complex outcomes. Using a data-driven approach on many distinct data elements may be necessary to describe this large set of outcomes and thereby robustly depict the nuanced differences among TBI patients' recovery. In this work, we develop a method for modeling large heterogeneous data types relevant to TBI. Our approach is geared toward the probabilistic representation of mixed continuous and discrete variables with missing values. The model is trained on a dataset encompassing a variety of data types, including demographics, blood-based biomarkers, and imaging findings. In addition, it includes a set of clinical outcome assessments at 3, 6, and 12 months post-injury. The model is used to stratify patients into distinct groups in an unsupervised learning setting. We use the model to infer outcomes using input data, and show that the collection of input data reduces uncertainty of outcomes over a baseline approach. In addition, we quantify the performance of a likelihood scoring technique that can be used to self-evaluate the extrapolation risk of prognosis on unseen patients.

We present a novel neural network architecture called AutoAtlas for fully unsupervised partitioning and representation learning of 3D brain Magnetic Resonance Imaging (MRI) volumes. AutoAtlas consists of two neural network components: one neural network to perform multi-label partitioning based on local texture in the volume, and a second neural network to compress the information contained within each partition. We train both of these components simultaneously by optimizing a loss function that is designed to promote accurate reconstruction of each partition, while encouraging spatially smooth and contiguous partitioning, and discouraging relatively small partitions. We show that the partitions adapt to the subject specific structural variations of brain tissue while consistently appearing at similar spatial locations across subjects. AutoAtlas also produces very low dimensional features that represent local texture of each partition. We demonstrate prediction of metadata associated with each subject using the derived feature representations and compare the results to prediction using features derived from FreeSurfer anatomical parcellation. Since our features are intrinsically linked to distinct partitions, we can then map values of interest, such as partition-specific feature importance scores onto the brain for visualization.

For our direct ink write (DIW) setup, hours of video are collected of fibrous ink that is printed from a translucent nozzle while parts are made. Due to sporadic misalignment of the fibers, clogs may arise that disrupt ink flow, resulting in a failed part. Without on-line monitoring, defective parts can only be identified by operators who witness clogs as or after they occur, requiring operators to continuously monitor the process to eliminate defects. In order to alleviate this, we aim to minimize the effect of clogging via automated process monitoring and rapid detection, thereby reducing labor costs, material loss, and proper identification of defective parts. In this paper, we propose applying a convolutional neural network (CNN) for single frame classification on images gathered from our DIW setup. We report a class average recall of 99.85% across clogged and unclogged classes, and average error of 1.64% when evaluated on new test video sequences, with a processing rate of 27.58 fps. Using class activation mapping, we can visualize image regions the CNN model identifies as salient for each class in performing its discriminative classification task.

The electrocardiogram (ECG) is an emerging biometric modality that has seen about 13 years of development in peer-reviewed literature, and as such deserves a systematic review and discussion of the associated methods and findings. In this paper, we review most of the techniques that have been applied to the use of the electrocardiogram for biometric recognition. In particular, we categorize the methodologies based on the features and the classification schemes. Finally, a comparative analysis of the authentication performance of a few of the ECG biometric systems is presented, using our inhouse database. The comparative study includes the cases where training and testing data come from the same and different sessions (days). The authentication results show that most of the algorithms that have been proposed for ECG-based biometrics perform well when the training and testing data come from the same session. However, when training and testing data come from different sessions, a performance degradation occurs. Multiple training sessions were incorporated to diminish the loss in performance. That notwithstanding, only a few of the proposed ECG recognition algorithms appear to be able to support performance improvement due to multiple training sessions. Only three of these algorithms produced equal error rates (EERs) in the single digits, including an EER of 5.5% using a method proposed by us.

A novel approach for remotely sensing mechanical cardiovascular activity for use as a biometric marker is proposed. Laser Doppler Vibrometry (LDV) is employed to sense vibrations on the surface of the skin above the carotid artery related to arterial wall movements associated with the central blood pressure pulse. Carotid LDV signals are recorded using noncontact methods and the resulting unobtrusiveness is a major benefit of this technique. Several recognition methods are proposed that use the temporal and/or spectral information in the signal to assess biometric performance both on an intrasession basis, and on an intersession basis where LDV measurements were acquired from the same subjects after delays ranging from one week to six months. A waveform decomposition method that utilizes principal component analysis is used to model the signal in the time domain. Authentication testing for this approach produces an equal-error rate of 0.5% for intrasession testing. However, performance degrades substantially for intersession testing, requiring a more robust approach to modeling. Improved performance is obtained using techniques based on time-frequency decomposition, incorporating a method for extracting informative components. Biometric fusion methods including data fusion and information fusion are applied to train models using data from multiple sessions. As currently implemented, this approach yields an intersession equal-error rate of 6.3%.