Constructing an IOT-based assistant service for recognizing vocal cord diseases

In this paper, we apply the Internet of Things (IoT) technology to construct the non-invasive examination, named IoT-based vocal cords (VC) disease inference system (i-VCD), to provide the disease inference assistant framework for physicians. The proposed i-VCD tracks patient’s voice recording during consulting by the IoT technology, analyzes the voice features, and outputs potential VC diseases. We evaluate several classification algorithms, including eXtreme gradient boosting (XGBoost), random forest, support vector machines, and artificial neural networks, to recognize the diseases based on the voice features. In the experiments, polyps, paralysis, and Reinke’s edema are considered as the target diseases, and two scenarios are proposed: the one-to-one model and the one-to-many model. In the one-to-one model, a classification algorithm is applied to recognize exactly one VC disease, while four diseases are evaluated together in the one-to-many model. The performance in the one-to-many model is worse than that in the one-to-one model because the sound features may overlap in various diseases. However, the one-to-many model is close to the clinical environment. The experiment results show that the i-VCD with XGBoost in the one-to-one model has 94%, 100%, and 100% for polyps, paralysis, and Reinke’s edema in accuracy, respectively. The accuracy is 93% in the one-to-many model, which outperforms related approaches. Moreover, i-VCD is also deployed in a cloud service so that the physicians can get the assistance of i-VCD easily. Eventually, i-VCD provides high performance in recognizing VC diseases in a non-invasive way and is helpful in clinical consulting.