In 2008, Amyotrophic lateral sclerosis (ALS) or Lou Gehrig's Disease became a presumptively compensable (service connected) disease as the Institute of Medicine (IOM) Committee stated an association between the development of ALS and military service. According to the IOM report, military service increases life risk of ALS by 1.5 fold. There are approximately 4,200 Veterans with ALS and roughly 1,000 new cases each year. At the Tampa VA, since 2007, there has been a consistent rise in the number of Veterans diagnosed and treated with ALS. Most physiological assessments that are commonly used to determine the functional status of patients with ALS require trained clinical personnel to administer and interpret the results. We propose to use automatic speech understanding and machine learning software (DESIPHER) to: identify speech pathologies and use them to predict other aspects of physiological degeneration associated with ALS (e.g., respiratory difficulty or inability to swallow), and ultimately improve speech recognition for those with speech impairments. We expect this to improve our ability to appropriately identify and intervene when Veterans with ALS are at risk of serious adverse medical issues such as respiratory failure and aspiration. We postulate that analyzing the overall divergence of (impaired) speech, from a normal baseline, will prove to be more robust and a better marker for involvement than others that have been proposed. Specific research questions to be addressed by this study are: (1) Is it possible to train a speech recognition system to adapt to increasingly more frequent language/speech errors of particular types, to produce an accurate textual transcript that would be readable by an ALS patient's caregiver or physician? (2) Are specific changes in physiological functioning; Forced Vital Capacity, tongue strength, speech velocity, weight (loss), aspiration risk, or psychological distress, reflected in different types of language/speech errors associated with ALS? By understanding how speech functioning correlates with the degree to which other biophysical functioning has degraded, it is possible to apply a new, non-invasive measure for assessing the functionality of an ALS patient. In addition, the features associated with speech degradation it is possible to adapt existing speech recognition software to a patient's speech as it evolves over time, so that the quality of life for patients may be improved through conversation with a computer. Respiratory failure is the main cause of morbidity and mortality in ALS patients. We expect that the method of analyzing speech will present an excellent biomarker for respiratory function, as there is an expected increase in pauses during speech due to the necessity of increase frequency of respirations, a decrease in loudness, and decreased overall velocity of speech. A second major cause of death is aspiration. As the articular muscles decline, we expect to note a decrease in the clarity of speech. Speech involvement often precedes swallowing involvement; thus, we expect that increasing speech divergence will indicate potential aspiration risk.
A disease called Amyotrophic lateral sclerosis (or ALS), which leads to difficulty swallowing, breathing, and movement, has been found to be higher for those serving in the military than in the general population. There are approximately 4,200 Veterans with ALS and roughly 1,000 new cases each year. When doctors attempt to determine the degree to which an ALS patient is suffering from the disease, they apply tests that are 'graded' by experts. However, this approach to testing patients may not be very accurate. Researchers aim to use a system called DESIPHER to 'listen' to ALS patients and find speech mistakes related to their condition. Researchers believe that, by detecting different types of errors, DESIPHER serves as a new kind of indicator of medical problems such as difficulty breathing or swallowing, without human 'grading'. This may also lead to a better electronic system for understanding ALS patients' speech, even with lots of errors.
