Lymph Node Metastases in Papillary Thyroid Carcinoma can be Predicted by a Convolutional Neural Network: a
Antoinette Esce1, Jordan P Redemann2, Garth T Olson1
1Department of Surgery, Division of Otolaryngology Head and Neck Surgery, University of New Mexico Health Sciences Center, Albuquerque, NM, USA.
The Annals of Otology, Rhinology, and Laryngology
|March 10, 2023
Summary
Artificial intelligence (AI) can predict papillary thyroid carcinoma (PTC) nodal metastases using primary tumor images. Multi-institutional data validated AI's accuracy, improving staging and treatment decisions for PTC patients.
Area of Science:
- Oncology
- Pathology
- Artificial Intelligence
- Medical Imaging
Background:
- Nodal metastases in papillary thyroid carcinoma (PTC) impact patient staging and treatment.
- Lymph node removal is not always performed during thyroidectomy.
- Previous studies show artificial intelligence (AI) can predict PTC nodal metastases from primary tumor histopathology.
Purpose of the Study:
- To validate the capability of AI in predicting nodal metastases in PTC using multi-institutional data.
- To assess the robustness of AI algorithms trained on histopathology for predicting lymph node involvement.
Main Methods:
- Conventional PTC cases from two academic institutions were analyzed.
- AI algorithms, specifically a convolutional neural network (CNN), were trained and tested on histopathology slides.
- Performance was evaluated using receiver operator characteristic curves and the Youden J statistic.
Main Results:
- A combined multi-institutional algorithm achieved an area under the curve (AUC) of 0.84.
- The best combined algorithm demonstrated 68% sensitivity and 91% specificity in predicting nodal metastases.
- Single-institution algorithms showed lower performance when tested on external data (AUC 0.64).
Conclusions:
- A CNN can generate accurate and robust algorithms for predicting PTC nodal metastases from primary tumor histopathology alone.
- AI models show promise for improving the prediction of lymph node involvement in PTC, even with diverse data sources.
