Automatic exposure control (AEC) for CT based on neural network-driven patient-specific real-time assessment of dose distributions and minimization of the effective dose
Final Report Abstract
The aim of this project was to enable rapid patient-specific dose calculations before and after the CT scan and to use them for a patient-specific risk minimizing tube current modulation (riskTCM). Modern CT scanners provide automatic exposure control (AEC) techniques, including automatic tube current selection, automatic tube voltage selection, and angular and longitudinal tube current modulation (TCM), to reduce radiation dose delivered to the patient while maintaining image quality. Today’s TCM implementations aim at minimizing the mAs-product as a simple but highly approximate surrogate for patient dose, therefore they will be denoted as mAsTCM in the following. Using mAsTCM has proven to reduce the total mAs value by up to 60%, depending on body region and scan protocol, compared to a scan with constant tube current (noTCM). However, the actual patient risk, for example in form of measures such as the effective dose Deff, is not accounted for. In order to be able to optimize for the effective dose Deff or any other biologically meaningful dose or risk measure, the organ doses have to be estimated before the actual CT scan. Therefore, machine learning approaches are essential in order for riskTCM to be applicable in clinical practice since conventional Monte-Carlo-based methods are far too slow. Based on the topogram, a coarse CT image has to be estimated using neural networks. Based on this, a deep learning (DL)-organ segmentation and a DL-dose distribution estimation for every projection angle need to be performed. Then, the estimated effective dose per view can be calculated and based on this a new patient risk minimizing TCM can be computed. All these steps were part of this project. The results were evaluated in a simulation study and compared against the common TCM algorithm mAsTCM and a constant current. Our project results show that all anatomical regions can benefit from riskTCM. A reduction of effective dose of up to 30% compared to mAsTCM can be demonstrated, depending on body region and tube voltage. The introduction into practice would be rather straightforward since only software and no hardware adaptations would be required.
Publications
-
A Novel CT Tube Current Modulation Technique That Minimizes Patient Risk. RSNA 2021.
L. Klein, C. Liu, J. Steidel, S. Sawall, A. Maier, M. Lell, J. Maier & M. Kachelrieß
-
Organ-specific vs. patient risk-specific tube current modulation in thorax CT scans covering the female breast. 7th International Conference on Image Formation in X-Ray Computed Tomography, 54. SPIE.
Klein, Laura; Enzmann, Lucia; Byl, Achim; Liu, Chang; Sawall, Stefan; Maier, Andreas; Maier, Joscha; Lell, Michael & Kachelrieß, Marc
-
Patient‐specific radiation risk‐based tube current modulation for diagnostic CT. Medical Physics, 49(7), 4391-4403.
Klein, Laura; Liu, Chang; Steidel, Jörg; Enzmann, Lucia; Knaup, Michael; Sawall, Stefan; Maier, Andreas; Lell, Michael; Maier, Joscha & Kachelrieß, Marc
-
Potential CT Radiation Dose Reduction to the Female Breast by a Novel Risk-Minimizing Tube Current Modulation. RSNA 2022.
L. Klein, E. Baader, A. Byl, C. Liu, S. Sawall, A. Maier, M. Lell, J. Maier & M. Kachelrieß
-
Real‐time estimation of patient‐specific dose distributions for medical CT using the deep dose estimation. Medical Physics, 49(4), 2259-2269.
Maier, Joscha; Klein, Laura; Eulig, Elias; Sawall, Stefan & Kachelrieß, Marc
-
Risk-minimizing tube current modulation (riskTCM) for CT - Potential dose reduction across different tube voltages. ECR 2022.
L. Klein, C. Liu, J. Steidel, L. Enzmann, S. Sawall, J. Maier, A. Maier, M. Lell & M. Kachelrieß
-
Whole-Body Multi-Organ Segmentation Using Anatomical Attention. 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), 1-5. IEEE.
Liu, Chang; Denzinger, Felix; Folle, Lukas; Qiu, Jingna; Klein, Laura; Maier, Joscha; Kachelrieβ, Marc; Lell, Micheal & Maier, Andreas
