Classifying the Tongue Contours for Assistive Devices

As artificial intelligence and machine learning become more prevalent, they are becoming an increasingly important part of our daily lives. This enabled human-machine interactions to be easier. Especially in the medical field, developing automatic detection systems to diagnose diseases and tumors. These systems classify affected areas, analyze symptoms, and develop assistive devices, among others. With these assistive devices, disabled people could continue living their regular lives much easier. The hybrid approach is performed to classify the ultrasound tongue contours in this paper for assistive devices. Geometric analysis is applied to precise modeling and understanding of the tongue's shape and movement, which is crucial for accurate classification. By capturing detailed structural information, it enhances the system's ability to make reliable assessments and improve assistive device performance. The dataset used in this paper is an open-source dataset. Noise reduction techniques like Spackle Reducing Anisotropic Diffusion and Non-local Means are applied to enhance the quality of ultrasound videos and make them more suitable for further analysis. Performance metrics for classifying tongue contours are presented. 87.80 percent accuracy is obtained and precision, sensitivity, specificity and F1 score calculated.