Capsule network based analysis of histopathological images of oral squamous cell carcinoma

Oral cancer is one of the most prevalent malignancy affecting oral cavity. Determining the correct type of oral cancer at the early stages is important in designing a detailed treatment plan and predicting the response of the patient to the treatment being adopted. A major challenge lies in the detection of oral cancer from histopathological images. In oral malignancy diagnosis, the main visual features are generally extracted from the architectural differences of epithelial layers and the appearance of keratin pearls. This paper proposes a new approach for classifying oral cancer using a deep learning technique known as capsule network. Dynamic routing and routing by agreement of capsule network makes it more robust for rotation and affine transformation of augmented oral dataset. This network's capability of handling pose, view and orientation makes it suitable for analysis of oral cancer histopathological images at an early stage. The performance of cross-validation indicate that the proposed methodology can efficiently classify the histopathological images of Oral Squamous Cell Carcinoma (OSCC) with 97.78% sensitivity, 96.92% specificity and 97.35% accuracy. (c) 2020 The Authors. Published by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Automated summary

Oral cancer is a common malignancy affecting the oral cavity. This paper presents a new approach for classifying oral cancer using a deep learning technique known as the capsule network. The network demonstrates high accuracy and sensitivity in classifying early-stage pathological images of oral cancer.