Related Experiment Video
Updated: Aug 7, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Deep unrolled regularization for limited-angle dual-panel positron emission mammography
Fernando Moncada-Gutiérrez1, Héctor Alva-Sánchez2, Mercedes Rodríguez-Villafuerte2
1Instituto de Física, Universidad Nacional Autónoma de México, Circuito de la Investigación Científica, Ciudad Universitaria, 04510, Mexico City, Mexico. moncadafer@estudiantes.fisica.unam.mx.
Physical and Engineering Sciences in Medicine
|August 6, 2026
Summary
This study introduces a deep learning method to fix image artifacts in dual-panel Positron Emission Mammography (PEM). The new technique significantly improves image quality, offering a promising advancement for breast cancer detection.
Area of Science:
- Medical Imaging
- Radiological Physics
- Artificial Intelligence in Medicine
Background:
- Dual-panel Positron Emission Mammography (PEM) offers superior resolution and sensitivity for breast imaging compared to conventional Positron Emission Tomography (PET).
- A key challenge in dual-panel PEM is the presence of limited-angle artifacts in cross-plane reconstructed images due to its specific acquisition geometry.
Purpose of the Study:
- To develop and evaluate a novel image reconstruction framework for dual-panel PEM that mitigates limited-angle artifacts.
- To enhance the diagnostic quality of dual-panel PEM images through deep learning-based regularization.
Main Methods:
- A U-Net deep learning model was trained to identify and correct artifacts in cross-plane slices using Monte Carlo (MC)-generated images from a ring-shaped dedicated breast PET as artifact-free ground truth.
- The trained deep learning model was integrated into the Forward-Backward Splitting Expectation-Maximization (FBSEM) algorithm for image reconstruction.
- Extensive data augmentation resulted in a training dataset of 3840 image pairs to improve model generalization.
Main Results:
- The proposed deep learning-enhanced reconstruction framework effectively reduced limited-angle artifacts in both MC-generated and experimental dual-panel PEM data.
- Quantitative analysis showed superior performance compared to OSEM and MAPEM across various image quality metrics, including noise, contrast, and resolution.
- The study presents the first application of deep learning-based regularization to address limited-angle artifacts in dual-panel PEM.
Conclusions:
- Deep unrolled regularization effectively mitigates limited-angle artifacts in dual-panel PEM imaging, significantly improving image reconstruction quality.
- This proof-of-concept study demonstrates the potential of AI-driven reconstruction for advancing breast-dedicated PET imaging.
- Further validation with anthropomorphic phantoms is necessary for potential clinical implementation.
