Laryngeal Image Dataset Automatic Annotation and Classification of Laryngeal Disease

Xinxin Xu,Xinyu Zhao,Weichen Xu,Jian Cao,Xing Zhang

DOI: https://doi.org/10.1109/iaecst60924.2023.10502675

2023-01-01

Abstract:Laryngeal disease is prevalent worldwide and has garnered increasing attention in the medical field due to its widespread occurrence. This paper focuses on addressing the early diagnosis of laryngeal disease, particularly addressing issues in laryngoscopy examination. A novel automatic annotation method for laryngeal datasets is proposed, leveraging the powerful Zero-Shot segmentation capability of the Segment Anything Model (SAM) and the inherent geometric features of the laryngeal data. The method employs a Dedicated High-quality annotation Generator (DHG) and a General Dataset Annotator (GDA) for automatic dataset annotation. By transforming an object classification dataset into an object detection dataset, the aim is to enhance the accuracy of the final diagnostic results. In the experiments, an object detection model is trained and tested on Laryngoscope8, a publicly available dataset of laryngeal disease (including seven pathological labels and one normal label), for symptom recognition. The experimental results demonstrate that our proposed method achieves competitive classification accuracy compared to other models, showcasing its effectiveness in addressing the challenges of early diagnosis of laryngeal disease.