An artificial intelligence (AI) ultrasound platform that incorporates multiple methods of machine learning can accurately predict thyroid malignancy as well as pathologic and genomic outcomes, according to data presented at the 2022 Multidisciplinary Head and Neck Cancers Symposium.1
Findings from the analysis of nearly 784 patients showed that the platform—which consists of radiomics, topologic data analysis, machine learning thyroid imaging reporting and data system (TI-RADS) features, and deep learning—predicted thyroid nodule malignancies with a 98.7% accuracy rate. In addition, tumor stage and BRAF V600E mutation status were predicted with an accuracy of 93% and 96%, respectively.
“We have developed an AI platform that examines ultrasound images and predicts with high accuracy whether a potentially problematic thyroid nodule is, in fact, cancerous,” said principal investigator Annie W. Chan, MD, Director of the Head and Neck Radiation Oncology Research Program at the Mass General Cancer Center. “If it is cancerous, we can further predict the tumor stage, the nodal stage, and the presence or absence of BRAF mutation.”
As Dr. Chan explained, thyroid cancer is one of the most rapidly increasing cancers in the United States, largely due to increased detection and improved diagnostics. If caught early, however, this disease is highly treatable, she added, and patients generally can expect to live a long time after treatment.
Background and Study Details
To train and validate the AI platform, Dr. Chan and colleagues obtained 1,346 thyroid nodule images through routine diagnostic ultrasound from 784 patients. The ultrasound images were divided into two data sets: one for internal training and validation and another for external validation. Samples obtained from fine-needle biopsy were used to confirm malignancy. The research confirmed pathologic staging and mutational status with operative reports and genomic sequencing, respectively.
Unlike the conventional AI approach, researchers combined four AI methods for this model, including radiomics, topologic data analysis, machine learning, and deep learning. Radiomics is a method that extracts numerous features from medical images using data-characterization algorithms (eg, how grade-level intensity varies across images), whereas topologic data analysis is an emergent technique that assesses the spatial relationship between data points in an image. A machine learning algorithm that used TI-RADS–defined ultrasound properties as machine learning features was also used. Finally, with deep learning, which requires a large data set, algorithms run data through multiple layers of an AI neural network to generate predictions.
Thyroid Malignancies Predicted With 98.7% Accuracy
Through these four methods, the AI platform accurately predicted 98.7% of thyroid nodule malignancies in the internal data set, reportedly outperforming individual AI modalities used alone. The model was also validated on the external data set, where it demonstrated a 93% accuracy for malignancy prediction. By comparison, the individual radiomics model predicted 89% of malignancies, the deep learning model achieved 87% accuracy, and topologic data analysis and machine learning with TI-RADS were accurate for 81% and 80% of the samples, respectively.
“By integrating different AI methods, we were able to capture more data while minimizing noise,” Dr. Chan observed. “This allows us to achieve a high level of accuracy in making predictions.”
The multimodal model was able to distinguish pathologic stage (93% accuracy for T stage, 89% for N stage, and 98% for extrathyroidal extension). In addition, the AI platform predicted the BRAF V600E mutation with 96% accuracy. As Dr. Chan explained, BRAF V600E is common in papillary thyroid carcinoma and is associated with a poor prognosis, but this type of thyroid cancer can be treated with targeted therapy.
“Our AI platform could present a low-cost, noninvasive option for screening, staging, and personalized treatment planning for the disease,” Dr. Chan concluded.
DISCLOSURE: Dr. Chan reported no conflicts of interest.
1. Paul R, Juliano A, Faquin W, et al: 2022 Multidisciplinary Head and Neck Cancers Symposium. Abstract 10. Presented February 25, 2022.
Alexander T. Pearson, MD, PhD, Assistant Professor of Medicine at the University of Chicago, commended the artificial intelligence (AI)-augmented, ultrasound-based platform for screening and staging of thyroid cancer.
“In this study, Dr. Chan and colleagues processed high-resolution ultrasound...