Related Experiment Video
Updated: Jan 5, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.2K
Mapping Driver Mutations to Histopathological Subtypes in Papillary Thyroid Carcinoma: Applying a Deep Convolutional
1Department of Genomic Medicine, University of Texas, MD Anderson Cancer Center, 1901 East Road, 3SCR5.4101, Houston, TX 77054, USA. peiling.tsou@gmail.com.
Journal of Clinical Medicine
|October 17, 2019
Summary
Deep learning models can predict papillary thyroid carcinoma (PTC) driver gene mutations (BRAF or RAS) using only histopathology images. This approach shows high accuracy and aids clinical decisions.
Area of Science:
- Oncology
- Computational Biology
- Pathology
Background:
- Papillary thyroid carcinoma (PTC) is the most common thyroid cancer subtype.
- Accurate biomarkers are crucial for risk stratification and guiding treatment in PTC.
- BRAF and RAS mutations are common drivers in PTC, occurring in approximately 50% and 10-15% of cases, respectively.
Purpose of the Study:
- To investigate the capability of deep learning, specifically convolutional neural networks (CNNs), to predict BRAF or RAS driver gene mutations in PTC using only histopathology images.
- To assess the performance of a CNN model trained on The Cancer Genome Atlas (TCGA) data for mutation prediction.
- To evaluate the correlation between image-based mutation classification and mRNA expression patterns.
Main Methods:
- A deep learning convolutional neural network (CNN) model, Google Inception v3, was trained on histopathology images from The Cancer Genome Atlas (TCGA).
- The model was designed to classify PTCs based on the presence of BRAF or RAS mutations.
- Model performance was evaluated using metrics such as the area under the curve (AUC) and accuracy on independent test sets, and correlated with mRNA expression data.
Main Results:
- The CNN model achieved high performance, with AUC values ranging from 0.878 to 0.951, comparable to other cancer type studies.
- On an independent testing subset, the model demonstrated 95.2% accuracy in predicting BRAF and RAS mutation classes.
- Image-based classification showed a strong correlation with mRNA expression patterns (Spearman correlation rho = 0.63 on validation data, rho = 0.79 on testing data).
Conclusions:
- Deep learning approaches hold significant potential for classifying cancer based on driver mutations using histopathology images.
- CNNs can effectively predict BRAF and RAS mutations in papillary thyroid carcinoma from image data alone.
- This image-based predictive capability could serve as a valuable tool to assist in clinical decision-making for PTC patients.
