Related Experiment Video
Updated: Sep 16, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
MRASM: A multiscale residual attention spatiotemporal model for breast tumor prediction
Liangliang Zhang1, Hang Zheng1, Jiawei Liu1
1School of Computer and Information, Anqing Normal University, Anqing 246133, People's Republic of China.
Background:
Spatial features and temporal features derived from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) are both useful for the prediction of tumor information. However, it remains unclear whether deep spatiotemporal features can improve the diagnostic performance of the model.
Purpose:
To improve the prediction performance of benign and malignant breast tumors by efficiently integrating deep spatial and temporal features from DCE-MRI.
Method:
A multiscale residual attention spatiotemporal model (MRASM) based on ResNet (2 + 1)D was proposed for the prediction of benign and malignant breast tumors. Compared to ResNet (2 + 1)D, the MRASM has three improvements, including multiscale, attention, and residual modules. Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) and accuracy. Statistical comparisons between the MRASM and traditional methods were performed using the DeLong test.
Results:
The MRASM method obtained a higher AUC of 0.9962 and accuracy of 0.9660 compared to original ResNet (2 + 1)D method (AUC = 0.9182 and accuracy = 0.8340) and traditional 3D CNN methods (best AUC = 0.8752 and accuracy = 0.8128).
Conclusion:
The proposed MRASM method is an efficient approach for the accurate prediction of benign and malignant breast tumors.
