Rule-based automatic diagnosis of thyroid nodules from intraoperative frozen sections using deep learning

作者：

Highlights：

• We develop an automated rule-based system to differentiate thyroid nodules based on deep learning techniques.

• We present a rule-based protocol to fuse patch predictions for the diagnosis of the thyroid frozen section slide, allowing for the interpretability of the proposed system.

• The automated diagnostic system obtains a precision of benign and malignant categories of thyroid nodules as 95.3% and 96.7%, respectively.

• The proposed method could facilitate pathologists in achieving an efficient and accurate diagnosis of thyroid nodules based on frozen sections.

摘要

•We develop an automated rule-based system to differentiate thyroid nodules based on deep learning techniques.•We present a rule-based protocol to fuse patch predictions for the diagnosis of the thyroid frozen section slide, allowing for the interpretability of the proposed system.•The automated diagnostic system obtains a precision of benign and malignant categories of thyroid nodules as 95.3% and 96.7%, respectively.•The proposed method could facilitate pathologists in achieving an efficient and accurate diagnosis of thyroid nodules based on frozen sections.

论文关键词：Thyroid nodule,Frozen section,Whole slide image,Deep learning,Rule-based protocol

论文评审过程：Received 27 June 2019, Revised 10 February 2020, Accepted 23 June 2020, Available online 9 August 2020, Version of Record 9 August 2020.

论文官网地址：https://doi.org/10.1016/j.artmed.2020.101918