Novel Chewing Cycle Approach for Peak Detection Algorithm of Chew Count Estimation

Chew count is a critical parameter in analyzing mastication signals, yet traditional methods of manual counting by trained clinicians are often labor-intensive and prone to errors. As a result, there has been a growing interest among researchers in developing automated methods for estimating chew count. This article reviews the existing approaches, evaluates their effectiveness, and proposes a new approach based on optimization technique. This work proposes a novel approach to chew count estimation using particle swarm optimization (PSO) combined with a peak detection algorithm. The chewing dataset comprises signals collected from 20 participants consuming eight different food types, with proximity sensors (PSs) detecting temporalis muscle activity. The peak detection algorithm identifies key signal features, while PSO optimizes the peak prominence and width parameters to minimize the mean absolute error (MAE) in chew count estimation. Two specific chewing cycle approaches were implemented: a participants-based (P) cycle and a participants-food type-based (PF) cycle. These approaches were compared to the traditional All (A) chewing cycle method, which evaluates chew count across the entire dataset in a single analysis. Results demonstrate that the PF method yields the lowest MAE at 1.25%, followed by the P method at 3.46%, and the A method at 4.26%. Moreover, the PF method required the least computational time at 8012.2 s, compared to 9392.0 s for the P method and 36621.4 s for the A.