CNN Ensembles for Nuclei Segmentation on Histological Images of OED

Abstract

Early diagnosis of potentially malignant disorders, such as oral epithelial dysplasia (OED), is the most reliable way to prevent oral cancer. Computational algorithms have been used as a tool to aid specialists in this process. In recent years, CNN-based methods have gained more attention due to their improved results in nuclei segmentation tasks. Despite these relevant results, achieving high segmentation accuracy remains a challenging task. In this paper, we propose an ensemble of segmentation models to improve the performance of nuclei segmentation in OED histopathology images. The proposed ensemble consists of four CNN segmentation models, which were combined using three ensemble strategies: simple averaging, weighted averaging and majority voting, achieved accuracy of 90.69%, 90.70% and 88.49%, respectively, when applied to OED images. The model's performance was also evaluated on three publicly available datasets and achieved comparable performance to state-of-the-art segmentation methods. These values indicate that the proposed ensemble methods can be used in medical image analysis applications.