End2end-ALARA: Approaching the ALARA Law in CT Imaging with End-to-end Learning

TL;DR

This work tackles reducing CT radiation exposure while preserving diagnostic quality by proposing End2end-ALARA, an end-to-end framework that jointly optimizes a dose modulation module and an image reconstruction module via a differentiable simulation and a constrained hinge loss. The method minimizes the delivered dose $D$ subject to a prescribed IQ level $q$, effectively enforcing $Q(R(p)) \ge q$ during training. A ResNet-18-based dose predictor maps prior patient information to $n_0 = e^d$, and a differentiable noise-injection simulation connects dose to projections, followed by FBP reconstruction and a Unet-based reconstructor; PSNR is used as the IQ metric with a target $q = 38.57$. Experiments on LoDoPaB-CT show End2end-ALARA achieves stable IQ across patients and requires lower dose than fixed-dose or conventional TCM to reach the same IQ, highlighting a practical pathway toward ALARA in CT imaging. The framework’s reliance on differentiable components and clinically meaningful IQ metrics suggests broad potential for personalized, dose-optimized CT workflows and downstream task training.

Abstract

Computed tomography (CT) examination poses radiation injury to patient. A consensus performing CT imaging is to make the radiation dose as low as reasonably achievable, i.e. the ALARA law. In this paper, we propose an end-to-end learning framework, named End2end-ALARA, that jointly optimizes dose modulation and image reconstruction to meet the goal of ALARA in CT imaging. End2end-ALARA works by building a dose modulation module and an image reconstruction module, connecting these modules with a differentiable simulation function, and optimizing the them with a constrained hinge loss function. The objective is to minimize radiation dose subject to a prescribed image quality (IQ) index. The results show that End2end-ALARA is able to preset personalized dose levels to gain a stable IQ level across patients, which may facilitate image-based diagnosis and downstream model training. Moreover, compared to fixed-dose and conventional dose modulation strategies, End2end-ALARA consumes lower dose to reach the same IQ level. Our study sheds light on a way of realizing the ALARA law in CT imaging.

Figures (4)

• Figure 1: The proposed End2end-ALARA framework exploring the way to realize the ALARA law in CT imaging. The key is to jointly optimize dose and reconstruction modules to minimize dose for every patient subject to a prescribed quality level $q$.
• Figure 2: Imaging results by different methods. For each method we show FBP-reconstructed and Unet-processed image on the left and right respectively. The proposed method automatically sets $n_0$ to acquire stable PSNRs across patients with various sizes and anatomies.
• Figure 3: Scatters of quality indices of 3552 cases in the test set by different methods. Comparing with the fixed-dose and conventional TCM methods, the proposed method produces images with quality stabled at the prescribed IQ level.
• Figure 4: The radiation dose consumed to reach the same IQ level.