Related Experiment Video
Updated: May 1, 2026
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
A deep learning framework for reconstructing Breast Amide Proton Transfer weighted imaging sequences from sparse
Qiuhui Yang1, Shu Su2, Tianyu Zhang3
1Faculty of Applied Sciences, Macao Polytechnic University, Macao Special Administrative Region of China; Guangxi Key Laboratory of Machine Vision and Intelligent Control, Wuzhou University, Wuzhou, China.
A new deep learning model reconstructs Amide Proton Transfer (APT) MRI data, significantly reducing scan times by 25% while maintaining image quality. This advancement in functional MRI could improve clinical applications by enabling faster protein metabolism quantification.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Biophysics
Background:
- Amide Proton Transfer (APT) MRI is a functional MRI technique for protein metabolism quantification.
- Clinical adoption of APT MRI is hindered by long acquisition times.
- Acquiring fewer frequency offset images to shorten scan times compromises quantification accuracy due to inadequate z-spectral fitting.
Purpose of the Study:
- To develop a deep learning model for reconstructing dense frequency offsets from sparse APT MRI data.
- To reduce APT MRI scanning time without compromising quantification accuracy.
- To improve the clinical applicability of APT MRI.
Main Methods:
- A deep learning model utilizing time-series convolution was developed to reconstruct dense frequency offsets from sparse APT MRI data.
- The model was trained to extract spatial and frequency features from the APT imaging sequence.
- A weighted layer was incorporated to assess the significance of individual frequency offsets.
Main Results:
- The proposed deep learning model achieved superior reconstruction performance compared to other seq2seq models.
- The model demonstrated a peak signal-to-noise ratio of 45.8 and a structural similarity index of 0.989 for tumor regions.
- Experimental results showed a 25% reduction in scanning time by reconstructing 29 dense frequency offsets from 21 sparse offsets.
Conclusions:
- The developed deep learning model effectively reconstructs dense frequency offsets from sparse APT MRI data, enabling a significant reduction in scanning time.
- This method offers potential guidance for optimizing APT imaging parameters and improving clinical workflow.
- The findings provide a valuable reference for clinicians utilizing APT MRI for protein metabolism quantification.
