Automatic discrimination between safe and unsafe swallowing using a reputation-based classifier

Abstract

BackgroundSwallowing accelerometry has been suggested as a potential non-invasive tool for bedside dysphagia screening. Various vibratory signal features and complementary measurement modalities have been put forth in the literature for the potential discrimination between safe and unsafe swallowing. To date, automatic classification of swallowing accelerometry has exclusively involved a single-axis of vibration although a second axis is known to contain additional information about the nature of the swallow. Furthermore, the only published attempt at automatic classification in adult patients has been based on a small sample of swallowing vibrations.MethodsIn this paper, a large corpus of dual-axis accelerometric signals were collected from 30 older adults (aged 65.47 ± 13.4 years, 15 male) referred to videofluoroscopic examination on the suspicion of dysphagia. We invoked a reputation-based classifier combination to automatically categorize the dual-axis accelerometric signals into safe and unsafe swallows, as labeled via videofluoroscopic review. From these participants, a total of 224 swallowing samples were obtained, 164 of which were labeled as unsafe swallows (swallows where the bolus entered the airway) and 60 as safe swallows. Three separate support vector machine (SVM) classifiers and eight different features were selected for classification.ResultsWith selected time, frequency and information theoretic features, the reputation-based algorithm distinguished between safe and unsafe swallowing with promising accuracy (80.48 ± 5.0%), high sensitivity (97.1 ± 2%) and modest specificity (64 ± 8.8%). Interpretation of the most discriminatory features revealed that in general, unsafe swallows had lower mean vibration amplitude and faster autocorrelation decay, suggestive of decreased hyoid excursion and compromised coordination, respectively. Further, owing to its performance-based weighting of component classifiers, the static reputation-based algorithm outperformed the democratic majority voting algorithm on this clinical data set.ConclusionGiven its computational efficiency and high sensitivity, reputation-based classification of dual-axis accelerometry ought to be considered in future developments of a point-of-care swallow assessment where clinical informatics are desired.

Anatomical Directional Dissimilarities in Tri-axial Swallowing Accelerometry Signals

“…In this study, 9 features in time, information-theoretic, frequency, and time-frequency domains were evaluated. Further, the practicality of these features had already been demonstrated in previous swallowing signal studies [22], [31], [32], [37], [50]. The computational details for each of these features are described in the following subsections.…”

“…On the other hand, some studies have extracted advanced features in time, frequency, and time-frequency domains from swallow vibrations [26], [31]–[34]. The results from past studies have shown the validity of the swallowing accelerometry method’s use of one or two directions of anterior-posterior (A-P) and superior-inferior (S-I), representing the forward and upward movements of hyolaryngeal structures during swallowing, in swallow classification and capturing swallowing difficulties [22], [26], [31], [32], [35], [36]. While the focus of most studies was on the assessing performance of different classification methods based on computed features, few studies have directly compared extracted features in two directions of A-P and S-I to find out about the dissimilarities between the swallowing vibrations in different anatomical directions [22], [36]–[38].…”

Reputation-Based Neural Network Combinations

“…Motivated by the above, we propose a novel algorithm for combining several classifiers based on their individual reputations, numerical weights that reflect each classifier's past performance. The algorithm is detailed in the next section and again is adapted from Nikjoo et al (2011).…”

“…The advantage of the reputation-based algorithm over the majority voting algorithm lies in the fact that the former has a higher probability of correct consensus and a faster rate of convergence to the peak probability of correct classification (Nikjoo et al, 2011).…”

“…In the following, we present notation that is used throughout the manuscript and detail the majority voting algorithm using this notation. The following presentation is adapted from (Nikjoo et al, 2011).…”