CAREER: Advanced data analytics and high-resolution cervical auscultation can accurately predict dysphagia

Overall: Dysphagia (swallowing disorders) causes 150,000 annual hospitalizations, increases pneumonia incidence, and adds 220,000 additional days to hospital admissions each year. Dysphagia risk is currently assessed via screening, and a failed screen then leads to the diagnostic gold-standard, the videofluoroscopic (VF) test. However, many patients who aspirate pass initial screens because they aspirate silently. The PI’s long-term research goal is to utilize computational approaches and instrumentation to nurture translation of innovative engineering research to clinically relevant solutions for dysphagia. Therefore, the research objective of this CAREER proposal is the advancement of fundamentally new data analytics tools to be used in conjunction with high-resolution cervical auscultation (HRCA - accelerometer and microphone recordings from the neck) in order to instrumentally screen for dysphagia. We have found that our HRCA screening device can detect pharyngeal vibrations caused by dysphagia. As of yet the current device cannot detect silent aspiration before patients are placed at risk. New fundamental advances in data analytics are needed, which is the main research contribution of the proposed program. The PI’s long-term educational goal is to translate advances in dysphagia data analytics into educational modules that will be disseminated to a wide clinical and engineering audience. Hence, the educational objective is to create interdisciplinary training opportunities for high school (e.g., educational modules to be distributed across the US), engineering undergraduate, and graduate students (e.g., an easily accessible open online course), while disseminating the results of this project to a clinical audience as well (e.g., grand rounds).
Intellectual Merit: This comprehensive CAREER program promotes expansion of both research and education in data analytics methods for dysphagia, with the goal of establishing the HRCA, coupled with data analytics, as a reliable screening tool for patients with swallowing difficulties. The proposed program is a major departure from the current efforts, as it focuses on the innovation of deep learning approaches for dysphagia. Deep learning represents state-of-the-art machine learning approaches based loosely on the processing of data in our brains, instead of learning from data features (traditional machine learning). Our transformative design addresses current technical obstacles: First, we aim to innovate deep learning approaches to model interactions between objective swallowing physiology observations from VF recordings and the co-occurring HRCA signals. These new deep learning approaches will be based on sequential Monte Carlo methods (to obtain interactions) and autoencoders (to find most compact models). Second, we will advance deep learning methods to predict the continuum of abnormal airway protection during swallowing. Our new approach is based on deep representation learning (to obtain concise multiclass representations) and multilevel deep classifiers (to differentiate multiple groups). The key transformative aspect of the proposed research consists of fundamental theoretical advancements of data analytics tools for swallowing difficulties, while translating the research into a clinically applicable tool.
Broader Impact: This CAREER proposal directly supports the President Obama’s Precision Medicine Initiative aimed at developing data-driven healthcare approaches. We will advance data analytics approaches that are applicable in other engineering (e.g., artificial intelligence, speech processing) and clinical fields (e.g., electrophysiology). Our proposed educational plan enhances the existing curricula by integrating new modules on emerging data analytics, and by utilizing a massive open online course platform and data sharing efforts. The PI will also actively recruit and encourage participation of women and underrepresented minority undergraduate and graduate students to the proposed project by expanding currently available opportunities at the University of Pittsburgh. In addition, the PI’s close collaboration with clinical partners offers the opportunity to transfer academic results into mainstream practices. Lastly, our work will further contribute to education through broad result dissemination via research publications, tutorials, workshops, clinical grand rounds, and technology transfer activities.