Unsupervised knowledge-transfer for learned image reconstruction

Abstract

Deep learning-based image reconstruction approaches have demonstrated impressive empirical performance in many imaging modalities. These approaches usually require a large amount of high-quality paired training data, which is often not available in medical imaging. To circumvent this issue we develop a novel unsupervised knowledge-transfer paradigm for learned reconstruction within a Bayesian framework. The proposed approach learns a reconstruction network in two phases. The first phase trains a reconstruction network with a set of ordered pairs comprising of ground truth images of ellipses and the corresponding simulated measurement data. The second phase fine-tunes the pretrained network to more realistic measurement data without supervision. By construction, the framework is capable of delivering predictive uncertainty information over the reconstructed image. We present extensive experimental results on low-dose and sparse-view computed tomography showing that the approach is competitive with several state-of-the-art supervised and unsupervised reconstruction techniques. Moreover, for test data distributed differently from the training data, the proposed framework can significantly improve reconstruction quality not only visually, but also quantitatively in terms of PSNR and SSIM, when compared with learned methods trained on the synthetic dataset only.

Keywords: Bayesian deep learning; computed tomography; image reconstruction; pretraining; unsupervised learning.

Figure 1.

(a) Schematic illustration of the overall iterative reconstructive process, and (b) the architecture of each sub-network.

Figure 2.

Representative ground truth images from Ellipses (left), FoamFanB (middle) and LoDoFanB (right) datasets. The window of the LoDoFanB dataset is set to a Hounsfield unit (HU) range ≈ [−1000, 400].

Figure 3.

Sparse-view reconstruction of the FoamFanB dataset along with a zoomed-in region indicated by a small square.

Figure 4.

Low-dose human chest CT reconstruction within the LoDoFanB dataset along with a zoomed-in region indicated by a small square. The window is set to a HU range of ≈[−1000, 400].

Figure 5.

Qualitative uncertainty analysis on the FoamFanB dataset. The pixel-wise absolute reconstruction error, (max-min normalised across low-dose and sparse-view CT settings) pixel-wise predictive uncertainty, and its decomposition into the aleatoric and epistemic constituent components for low-dose and sparse-view CT obtained by BDGD and BDGD + UKT.

Figure 6.

Qualitative uncertainty analysis on the LoDoFanB dataset. The pixel-wise absolute reconstruction error, (max-min normalised across low-dose and sparse-view CT settings) pixel-wise predictive uncertainty, and its decomposition into the aleatoric and epistemic constituent components for low-dose and sparse-view CT obtained by BDGD and BDGD + UKT.