A CNN-LSTM Hybrid Model for Estimation of Swallowing Based on Larynx sEMG

Total laryngectomy (TL), a surgical intervention for laryngeal pathologies, necessitates the complete excision of the larynx. However, patients have difficulties communicating daily, coughing, and swallowing after surgery. The artificial bioengineered larynx (ABL) is one of the ways to achieve the function by substituting the original larynx. With the ever-growing patients suffering from the larynx diseases and increasing demand for restoration or substitution of the larynx function after the radical surgeries, this paper introduces a method to estimate the 3 kinds of swallowing action based on larynx sEMG signals from submental, intercostal and diaphragm muscles by using a type of hybrid deep learning model, CNN-LSTM, for the artificial larynx. The algorithm proposed in this paper achieves an average RMSE of $\mathbf{0. 0 0 0 3 5 6 4 1}$ and an MAE of 0.00026872. Upon evaluating the four types of conventional classifiers, it has been determined that the model exhibits superior performance across all programs. The experimental results suggest the effectiveness of the proposed model.