Artificial intelligence platform combines multiple methods, including radiomics, topological data analysis, deep learning, and machine learning
WEDNESDAY, March 16, 2022 (HealthDay News) -- An artificial intelligence (AI) platform may detect thyroid malignancy and predict pathological and genomic outcomes through analysis of images from routine ultrasonography, according to a study presented at the 2022 Multidisciplinary Head and Neck Cancers Symposium, held from Feb. 24 to 26 in Phoenix.
Rahul Paul, Ph.D., from Massachusetts General Hospital in Boston, and colleagues developed a multimodal AI ultrasound platform that uses radiomics, topological data analysis (TDA), machine learning of Thyroid Imaging Reporting & Data System (TI-RADS)-defined ultrasound properties, and deep learning (DL) to predict malignancy and pathological outcome in patients with papillary thyroid cancer. Analysis included an internal training (103 malignant and 259 benign nodules), an internal validation (51 malignant and 98 benign nodules), and an external validation (270 malignant and 50 benign nodules) dataset.
The researchers found that for malignancy prediction, radiomics, TDA, TI-RADS, and deep learning model achieved an accuracy of 88.7, 81.5, 80.0, and 87.4 percent, respectively. The accuracy of the multimodal AI platform was 98.7 percent, which is significantly better than any individual model. The multimodal model achieved 91.4 percent accuracy for malignancy prediction in the external validation dataset. Using the multimodal model that consists of radiomics, TDA, and TI-RADS, the accuracy was 93 percent for tumor stage, 89 percent for nodal stage, and 98 percent for extrathyroidal extension. Accuracy was 96 percent for predicting BRAF mutation when employing a fivefold cross validation.
"This low-cost reproducible approach would allow developing of personalized treatment planning, especially in countries where resources are limited," the authors write.
Press Release
More Information