Neural networks in computer-aided diagnosis classification of dysphagic patients

Dysphagia is a swallowing disorder observed in head injury and stroke patients. Impairment of the oral and pharyngeal phases of the swallowing mechanism often leads to aspiration. Clinically, it is essential to estimate the risk of aspiration in dysphagic patients so that safe feeding protocols can be established. The current clinical procedures of diagnosis are qualitative and involve videofluorography and bedside clinical evaluation. Videofluorography exposes the patient to X-ray radiation and therefore cannot be performed on a daily basis. Reddy, et al. [1–3], have developed non-invasive biomechanical techniques to quantify oral and pharyngeal phases. The biomechanical parameters measured to characterize oral phase were the lateral tongue thrust, the forward tongue thrust, the tongue elevation force and the lip pulling force. For the pharyngeal phase, accelerations of the throat were measured by placing ultraminiature accelerometers on the skin at the levels of cricoid and thyroid cartilages, and the swallow suction pressure was measured with a catheter placed at the base of the tongue. Thomas, et al.[4], have used these biomechanical parameters to classify dysphagic patients into four categories of risk of aspiration. The purpose of the present investigation was to develop and clinically evaluate neural network models for classification of patients based on the biomechanical measurements obtained from them.