An Improved Classification Method of Ultrasonic Thyroid Standard Planes Based on Bayesian Optimization of Multi-Feature Parameters for Portable Application

In recent years, computer aided technology is increasingly applied to the classification of thyroid ultrasonic planes for calculating the thyroid volume, and then the disease can be diagnosed by its volume. However, these methods requiring greater computational power are not suitable for the portable ultrasound system with limited performance. Based on Bayesian optimization of multi-feature parameters, an improved classification method of ultrasonic thyroid planes is proposed in this paper for low computational complexity with a high classification accuracy. Local Binary Patterns (LBP) feature and Gray Level Co-occurrence Matrix (GLCM) feature are selected by the analysis of image texture information, and then the combined feature is utilized to investigate the effectiveness of different classifiers. Furthermore, Bayesian optimization is employed to obtain the optimal parameters of the combined feature for improving classification results, and the accuracy can reach up to 96.81%. The results clearly illustrate that the improved method is effective in classifying the ultrasonic thyroid planes under a low computational complexity.