Deep Learning in Thyroid Ultrasonography to Predict Tumor Recurrence in Thyroid Cancers

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Thyroid cancer diagnostics

Background:

• Thyroid cancer recurrence poses a significant clinical challenge.
• Accurate prediction of recurrence is crucial for patient management.
• Preoperative ultrasonography (US) is a key imaging modality for thyroid nodules.

Purpose of the Study:

• To evaluate a deep learning (DL) model for predicting thyroid tumor recurrence.
• To assess the efficacy of DL in analyzing preoperative US images for recurrence prediction.

Main Methods:

• A dataset of 898 US images from 229 thyroid cancer patients was created.
• A convolutional neural network (CNN) was developed using Python and Keras.
• The DL model's predictive performance was evaluated using receiver operating characteristic (ROC) analysis.

Main Results:

• Tumor recurrence was observed in 21.4% of patients.
• Tumor size and multifocality were significant predictors of recurrence.
• The DL model achieved a high overall area under the curve (AUC) of 0.9 ± 0.06 for predicting recurrence.

Conclusions:

• Deep learning analysis of US images can predict thyroid cancer recurrence.
• The developed DL model demonstrates strong potential for clinical application in managing thyroid cancer.