Background

In imaging modalities that use ionizing radiation, such as X-ray computed tomography (CT) or positron emission tomography (PET), photon scattering decreases image contrast and impedes image quantification. Therefore, scatter correction is essential, but accurate scatter estimation is often prohibitively time-consuming. Machine-learning can be employed for real-time, highly accurate scatter estimation (Deep Scatter Estimation, DSE), but so far, accuracy is limited by the use of measured photons in only a single energy bin.

Technology Overview

This invention is based on the assumption that large-angle, low-energy scatter (which is often discarded as much as possible) contains useful information about low-angle, high-energy scatter (which shall be estimated and removed from the measured data) that can be leveraged by machine learning. The invention uses energy measurements to discern measured photons by energy bin. Multiple raw-data images can be formed from raw data acquired in multiple energy bins, and used for machine-learning approaches such as convolutional neural networks (compare ).

Stage of Development

Deep scatter estimation has been successfully demonstrated in clinical X-ray CT [1] and clinical PET imaging [2]. Employing additional energy measurements is currently under investigation at the DKFZ.

Further Details

1. Maier, Kachelrieß et al. Med Phys 2018. https://doi.org/10.1002/mp.13274
2. Berker, Kachelrieß et al. IEEE NSS/MIC 2018. https://goo.gl/eCSeJC

Benefits

• Real-time scatter estimation, enabled by machine learning
• Successful demonstrations in X-ray computed tomography and positron emission tomography
• Improved accuracy promised by using energy measurements

Applications

The invention can be used in any industrial or medical imaging modality which uses ionizing radiation and suffers from scattered photons, and which can differentiate measured photons by their energy (e.g., positron emission tomography, single photon emission computed tomography, energy-selective X-ray computed tomography).