Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (40 results)

  • Thesis
  • Peer Reviewed
UCLA Electronic Theses and Dissertations (2015)

The ability to communicate using language is a fundamental human function. When this ability is compromised, as it can be in neuromuscular diseases such as amyotrophic lateral sclerosis (ALS) and brainstem strokes, patients stand to lose a significant source of functional independence. Brain-computer interface (BCI) systems help restore communication to these "locked-in" patients, usually relying on P300 evoked response potentials (ERPs) to identify a target character among repetitive serial presentation of possible characters. While the so-called "P300 speller" was first described over 25 years ago, little overall progress has been made with respect to clinical implementation, with major system limitations related to practicality, speed, and accuracy. This work addresses these concerns by using machine learning techniques to optimize the system design, accelerate the character selection process, and integrate natural language domain knowledge into the classifier. This effort has involved several different projects, including selecting the optimal electrode positions using Gibbs sampling, performing unsupervised training with the Baum-Welch algorithm, and incorporating prior language knowledge using particle filtering. The result is an online system requiring only four electrodes that allows users to communicate at an average bit rate that is 75% higher than when using standard methods. These improvements can help to make the P300 speller system a more viable solution for "locked-in" patients, leading to increased functional independence and improved quality of life.

    Cover page: Toward real-time communication using brain-computer interface systems
    • Article
    • Peer Reviewed
    UCLA Previously Published Works (2023)
    Medical natural language processing (NLP) systems are a key enabling technology for transforming Big Data from clinical report repositories to information used to support disease models and validate intervention methods. However, current medical NLP systems fall considerably short when faced with the task of logically interpreting clinical text. In this paper, we describe a framework inspired by mechanisms of human cognition in an attempt to jump the NLP performance curve. The design centers on a hierarchical semantic compositional model (HSCM), which provides an internal substrate for guiding the interpretation process. The paper describes insights from four key cognitive aspects: semantic memory, semantic composition, semantic activation, and hierarchical predictive coding. We discuss the design of a generative semantic model and an associated semantic parser used to transform a free-text sentence into a logical representation of its meaning. The paper discusses supportive and antagonistic arguments for the key features of the architecture as a long-term foundational framework.
      Cover page: Design considerations for a hierarchical semantic compositional framework for medical natural language understanding.
      • Thesis
      • Peer Reviewed
      UCLA Electronic Theses and Dissertations (2020)

      Recent development in electronic medical devices or systems has realized the effective collection and documentation of patients’ health in real time. To date, the potential clinical impact of this healthcare data has not been fully realized. Specifically, patients’ health data is heterogenous and sparse in nature, as it is composed of various modalities and is collected on different scales. In addition, processing this data efficiently in a temporal manner to take advantage of its sequential structure remains a barrier for medical records. This dissertation attempts to overcome these challenges by developing machine learning models to classify patient reported outcome (PRO) scores from activity tracker data and predict depression diagnoses based on data from patients’ historical electronic health records (EHR). A temporal model based on hidden Markov models (HMM) is first proposed to classify PRO scores in various categories from human vital signs collected from Fitbit activity trackers. This approach is able to combine various vital signs on difference scales in a single model that tracks changes in PRO scores over time. Second, several end-to-end machine learning models were built to aggregate multimodal EHR data in a single model. A novel hierarchical embedding method achieved superior performance for predicting depression diagnosis, which lays a foundation for addressing the heterogeneity and sparsity of EHR data. Third, an innovative bidirectional sequence learning model with a transformer architecture was developed for representation learning on high dimensional EHR data, demonstrating significantly improved performance over the traditional forward-only method. Finally, methods to improve the interpretability of the aforementioned models have been developed, which is a critical step before clinical deployment. Relative feature importance factors are determined for each vital sign collected from the Fitbit and attention weights are found for each data modality in the sequential EHR data. Extensive experiments and results have demonstrated the effectiveness of these proposed methods. This dissertation provides methodologies that advance modeling and understanding of digital health datasets, which lays the foundation to construct clinical decision support systems in this domain which could potentially lead to early disease detection and intervention.

        Cover page: Advancing Temporal Modeling and Heterogeneous Data Analysis for Digital Health
        • Article
        • Peer Reviewed
        UCLA Previously Published Works (2016)
        Probabilistic topic models provide an unsupervised method for analyzing unstructured text, which have the potential to be integrated into clinical automatic summarization systems. Clinical documents are accompanied by metadata in a patient's medical history and frequently contains multiword concepts that can be valuable for accurately interpreting the included text. While existing methods have attempted to address these problems individually, we present a unified model for free-text clinical documents that integrates contextual patient- and document-level data, and discovers multi-word concepts. In the proposed model, phrases are represented by chained n-grams and a Dirichlet hyper-parameter is weighted by both document-level and patient-level context. This method and three other Latent Dirichlet allocation models were fit to a large collection of clinical reports. Examples of resulting topics demonstrate the results of the new model and the quality of the representations are evaluated using empirical log likelihood. The proposed model was able to create informative prior probabilities based on patient and document information, and captured phrases that represented various clinical concepts. The representation using the proposed model had a significantly higher empirical log likelihood than the compared methods. Integrating document metadata and capturing phrases in clinical text greatly improves the topic representation of clinical documents. The resulting clinically informative topics may effectively serve as the basis for an automatic summarization system for clinical reports.
          Cover page: Using phrases and document metadata to improve topic modeling of clinical reports.
          • Article
          • Peer Reviewed
          UCLA Previously Published Works (2014)
          Despite the growing ubiquity of data in the medical domain, it remains difficult to apply results from experimental and observational studies to additional populations suffering from the same disease. Many methods are employed for testing internal validity; yet limited effort is made in testing generalizability, or external validity. The development of disease models often suffers from this lack of validity testing and trained models frequently have worse performance on different populations, rendering them ineffective. In this work, we discuss the use of transportability theory, a causal graphical model examination, as a mechanism for determining what elements of a data resource can be shared or moved between a source and target population. A simplified Bayesian model of glioblastoma multiforme serves as the example for discussion and preliminary analysis. Examination over data collection hospitals from the TCGA dataset demonstrated improvement of prediction in a transported model over a baseline model.
            Cover page: Motivating the additional use of external validity: examining transportability in a model of glioblastoma multiforme.
            • Article
            • Peer Reviewed
            UCLA Previously Published Works (2021)
            Advancements in machine learning algorithms have had a beneficial impact on representation learning, classification, and prediction models built using electronic health record (EHR) data. Effort has been put both on increasing models' overall performance as well as improving their interpretability, particularly regarding the decision-making process. In this study, we present a temporal deep learning model to perform bidirectional representation learning on EHR sequences with a transformer architecture to predict future diagnosis of depression. This model is able to aggregate five heterogenous and high-dimensional data sources from the EHR and process them in a temporal manner for chronic disease prediction at various prediction windows. We applied the current trend of pretraining and fine-tuning on EHR data to outperform the current state-of-the-art in chronic disease prediction, and to demonstrate the underlying relation between EHR codes in the sequence. The model generated the highest increases of precision-recall area under the curve (PRAUC) from 0.70 to 0.76 in depression prediction compared to the best baseline model. Furthermore, the self-attention weights in each sequence quantitatively demonstrated the inner relationship between various codes, which improved the model's interpretability. These results demonstrate the model's ability to utilize heterogeneous EHR data to predict depression while achieving high accuracy and interpretability, which may facilitate constructing clinical decision support systems in the future for chronic disease screening and early detection.
              Cover page: Bidirectional Representation Learning From Transformers Using Multimodal Electronic Health Record Data to Predict Depression
              • Article
              • Peer Reviewed
              UCLA Previously Published Works (2016)

              Background and objective

              Probabilistic topic models provide an unsupervised method for analyzing unstructured text. These models discover semantically coherent combinations of words (topics) that could be integrated in a clinical automatic summarization system for primary care physicians performing chart review. However, the human interpretability of topics discovered from clinical reports is unknown. Our objective is to assess the coherence of topics and their ability to represent the contents of clinical reports from a primary care physician's point of view.

              Methods

              Three latent Dirichlet allocation models (50 topics, 100 topics, and 150 topics) were fit to a large collection of clinical reports. Topics were manually evaluated by primary care physicians and graduate students. Wilcoxon Signed-Rank Tests for Paired Samples were used to evaluate differences between different topic models, while differences in performance between students and primary care physicians (PCPs) were tested using Mann-Whitney U tests for each of the tasks.

              Results

              While the 150-topic model produced the best log likelihood, participants were most accurate at identifying words that did not belong in topics learned by the 100-topic model, suggesting that 100 topics provides better relative granularity of discovered semantic themes for the data set used in this study. Models were comparable in their ability to represent the contents of documents. Primary care physicians significantly outperformed students in both tasks.

              Conclusion

              This work establishes a baseline of interpretability for topic models trained with clinical reports, and provides insights on the appropriateness of using topic models for informatics applications. Our results indicate that PCPs find discovered topics more coherent and representative of clinical reports relative to students, warranting further research into their use for automatic summarization.
                Cover page: Evaluating topic model interpretability from a primary care physician perspective
                • Article
                • Peer Reviewed
                UCLA Previously Published Works (2021)
                Recent developments in machine learning algorithms have enabled models to exhibit impressive performance in healthcare tasks using electronic health record (EHR) data. However, the heterogeneous nature and sparsity of EHR data remains challenging. In this work, we present a model that utilizes heterogeneous data and addresses sparsity by representing diagnoses, procedures, and medication codes with temporal Hierarchical Clinical Embeddings combined with Topic modeling (HCET) on clinical notes. HCET aggregates various categories of EHR data and learns inherent structure based on hospital visits for an individual patient. We demonstrate the potential of the approach in the task of predicting depression at various time points prior to a clinical diagnosis. We found that HCET outperformed all baseline methods with a highest improvement of 0.07 in precision-recall area under the curve (PRAUC). Furthermore, applying attention weights across EHR data modalities significantly improved the performance as well as the model's interpretability by revealing the relative weight for each data modality. Our results demonstrate the model's ability to utilize heterogeneous EHR information to predict depression, which may have future implications for screening and early detection.
                  Cover page: HCET: Hierarchical Clinical Embedding With Topic Modeling on Electronic Health Records for Predicting Future Depression
                  • Article
                  • Peer Reviewed
                  UCLA Previously Published Works (2013)
                  Brain-computer interface (BCI) systems are a promising means for restoring communication to patients suffering from "locked-in" syndrome. Research to improve system performance primarily focuses on means to overcome the low signal to noise ratio of electroencephalogric (EEG) recordings. However, the literature and methods are difficult to compare due to the array of evaluation metrics and assumptions underlying them, including that: 1) all characters are equally probable, 2) character selection is memoryless, and 3) errors occur completely at random. The standardization of evaluation metrics that more accurately reflect the amount of information contained in BCI language output is critical to make progress. We present a mutual information-based metric that incorporates prior information and a model of systematic errors. The parameters of a system used in one study were re-optimized, showing that the metric used in optimization significantly affects the parameter values chosen and the resulting system performance. The results of 11 BCI communication studies were then evaluated using different metrics, including those previously used in BCI literature and the newly advocated metric. Six studies' results varied based on the metric used for evaluation and the proposed metric produced results that differed from those originally published in two of the studies. Standardizing metrics to accurately reflect the rate of information transmission is critical to properly evaluate and compare BCI communication systems and advance the field in an unbiased manner.
                    Cover page: Evaluating True BCI Communication Rate through Mutual Information and Language Models
                    • Article
                    • Peer Reviewed
                    UCLA Previously Published Works (2014)
                    The P300 speller is a common brain-computer interface (BCI) application designed to communicate language by detecting event related potentials in a subject's electroencephalogram (EEG) signal. Information about the structure of natural language can be valuable for BCI communication systems, but few attempts have been made to incorporate this domain knowledge into the classifier. In this study, we treat BCI communication as a hidden Markov model (HMM) where hidden states are target characters and the EEG signal is the visible output. Using the Viterbi algorithm, language information can be incorporated in classification and errors can be corrected automatically. This method was first evaluated offline on a dataset of 15 healthy subjects who had a significant increase in bit rate from a previously published naïve Bayes approach and an average 32% increase from standard classification with dynamic stopping. An online pilot study of five healthy subjects verified these results as the average bit rate achieved using the HMM method was significantly higher than that using the naïve Bayes and standard methods. These findings strongly support the integration of domain-specific knowledge into BCI classification to improve system performance and accuracy.
                      Cover page: Integrating Language Information With a Hidden Markov Model to Improve Communication Rate in the P300 Speller
                      Top