Papillary Thyroid Carcinoma Semantic Segmentation Using Multi-Scale Adaptive Convolutional Network With Dual Decoders

Abstract

Papillary thyroid carcinoma (PTC) is the most common type of thyroid cancer. Currently, pathologists diagnose the PTC by interpreting their nuclei. However, the existing diagnosis method is difficult to interpret, especially for the cases falling in the borderline zones. According to the advances in artificial intelligence (AI) technology, semantic segmentation is used to support medical personnel. This study proposes Multi-scale Adaptive Convolutional Network with DBUNet (MSAC-DBUNet) and Multi-scale Adaptive Convolutional Network with Dual Decoders (MSAC-DD), which are more accurate and faster than the traditional networks. The experimental result shows that MSAC-DBUNet achieves good PTC segmentation outcome, and MSAC-DD provides a comparable score while reducing computation time and GPU usage compared to MSAC-DBUNet. It can be used for further studies to support the pathology practice.

Keywords

• Papillary thyroid carcinoma
• histopathology image
• pathology
• deep learning
• convolution neural network
• semantic segmentation