AI model can aid in thyroid cancer screening, staging and treatment planning

Obtain permission

By Matthew Stenger

Published: 03/03/2022 12:26:00

Last update: 03/03/2022 13:18:18

A new study has found that an artificial intelligence (AI) model incorporating several machine learning methods accurately detects thyroid cancer and predicts pathological and genomic outcomes through analysis of routine ultrasound images. The AI ​​model could present a low-cost, non-invasive option for disease screening, staging, and personalized treatment planning. The study results were presented by Chan et al at the 2022 Multidisciplinary Head and Neck Cancer Symposium (Summary 10).

“Thyroid cancer is one of the fastest growing cancers in the United States, thanks in large part to increased detection and improved diagnostics. We developed an AI platform that would examine images ultrasound scans and would predict with great accuracy whether a potentially problematic thyroid nodule is, in fact, cancerous.If it is cancerous, we can further predict the tumor stage, nodal stage, and the presence or absence of BRAF mutation,” said the lead author Annie Chan, MD, director of the head and neck radiation oncology research program at the Massachusetts General Cancer Center. “If detected early, this disease is highly treatable and patients can generally expect to live a long time after treatment.”

Creation and Performance of the Platform

To train and validate the AI ​​platform, the researchers obtained 1,346 images of thyroid nodules by routine diagnostic ultrasound from 784 patients. The ultrasound images were split into two data sets: one for internal training and validation and one for external validation. Malignancy was confirmed with specimens obtained by fine needle biopsy. Pathological staging and mutation status were confirmed by operative reports and genomic sequencing, respectively.

Unlike the conventional AI approach, the researchers combined multiple AI methods for the model, including radiomics, which extracts a large number of quantitative features from the images; topological data analysis (TDA), which assesses the spatial relationship between data points in images; deep learning, where algorithms run data through multiple layers of an AI neural network to generate predictions; and machine learning (ML), in which an algorithm uses ultrasound properties defined by Thyroid Imaging Reporting and Data System (TI-RADS) as machine learning capabilities.

“By integrating different AI methods, we were able to capture more data while minimizing noise. This allows us to achieve a high level of accuracy in making predictions,” Dr. Chan said.

A multimodal platform using these four methods accurately predicted 98.7% of thyroid nodule malignancies in the internal dataset, significantly outperforming individual AI modalities used alone. In comparison, the individual radiomics model predicted 89% of malignancies (P < 0.001 compared to the multimodal platform), the deep learning model achieved an accuracy of 87% (P = 0.002), and TDA and (ML)TI-RADS were accurate for 81% and 80% of samples, respectively (both P < .001). On the external validation dataset, the model was 93% accurate in predicting malignancy.

A multimodal model including radiomics, TDA and (ML)TI-RADS was also able to distinguish the pathological stage (93% accuracy for T stage, 89% for N stage and 98% for extrathyroidal extension). Moreover, the model identified BRAF Mutation V600E with 96% accuracy.

Disclosure: For full disclosures from the study authors, visit

The content of this article has not been reviewed by the American Society of Clinical Oncology, Inc. (ASCO®) and does not necessarily reflect the views and opinions of ASCO®.