Artificial intelligence to predict the BRAFV600E mutation in patients with thyroid cancer

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• The BRAFV600E mutation is a common genetic alteration in thyroid cancer, influencing treatment decisions and prognosis.
• Accurate prediction of this mutation can aid in personalized treatment strategies and patient management.

Purpose of the Study:

• To evaluate the efficacy of a deep learning-based computer-aided diagnosis (CAD) program utilizing neck ultrasound (US) images in predicting the BRAFV600E mutation in thyroid cancer.
• To identify independent predictive factors for the BRAFV600E mutation through multivariate analysis.

Main Methods:

• A retrospective study included 469 thyroid cancer cases from 469 patients.
• A deep convolutional neural network (CNN)-based CAD program assessed US images, providing malignancy risk scores (CAD values).
• Logistic regression analyses, including patient demographics, ACR TIRADS, and CAD values, were performed to identify predictive factors for the BRAFV600E mutation.

Main Results:

• The BRAFV600E mutation was present in 81% of patients.
• Multivariate analysis identified older age, smaller nodule size, and higher CAD values as significant predictors of the BRAFV600E mutation.
• The CAD value achieved an AUC of 0.646, while the multivariable model achieved an AUC of 0.706, indicating superior performance.

Conclusions:

• Deep learning-based CAD applied to thyroid US images shows potential in predicting the BRAFV600E mutation in thyroid cancer.
• Further multi-center validation studies are recommended to confirm these findings and enhance clinical applicability.