Abstract
Objective histopathological judgment is very important to determine the treatment and prognosis of oral squamous cell carcinoma. However, there are many problems in clinical practice, such as the chronic shortage of pathologists and disagreements among evaluators. Therefore, there are great expectations for the development of algorithms to support pathological diagnosis using machine learning.
Intercellular bridges, a histological feature of normal oral squamous epithelium, are known to disappear at poor differentiation in oral squamous cell carcinoma. However, the structures are extremely narrow and difficult for pathologists to distinguish and evaluate from the entire pathological tissue.
We evaluated the clinicopathology of intercellular bridges in oral squamous cell carcinoma. In addition, to provide an objective measure for this evaluation, we attempted automated segmentation of intercellular bridges using machine learning. As an area of interest, intercellular bridges in the infiltration depth region were selected using whole slide imaging of HE specimens of primary oral squamous cell carcinoma cases.
In the classification and evaluation of intercellular bridges, we found a close association between loss of intercellular bridges and poor prognosis. We were also able to develop an automated segmentation of those intercellular bridges. However, there are still many unstable aspects of automatic segmentation using machine learning, and there is still room for improvement for clinical application.
