Anomaly Detection and Artifact Recovery in PET Attenuation-Correction Images Using the Likelihood Function

Abstract

In dual modality PET/CT, CT data are used to generate the attenuation correction applied in the reconstruction of the PET emission image. This requires converting the CT image into a 511-keV attenuation map. Algorithms for making this transformation require assumptions about the makeup of material within the patient. Anomalous material such as contrast agent administered to enhance the CT scan confounds conversion algorithms and has been observed to result in inaccuracies, i.e., inconsistencies with the true 511-keV attenuation present at the time of the PET emission scan. These attenuation artifacts carry through to the final attenuation-corrected PET emission image and can resemble diseased tissue. We propose an approach to correcting this problem that employs the attenuation information carried by the PET emission data. A likelihood-based algorithm for identifying and correcting of contrast is presented and tested. The algorithm exploits the fact that contrast artifacts manifest as too-high attenuation values in an otherwise high quality attenuation image. In a separate study, the performance of the loglikelihood as an objective-function component of a detection/correction algorithm, independent of any particular algorithm was mapped out for several imaging scenarios as a function of statistical noise. Both the full algorithm and the loglikelihood performed well in studies with simulated data. Additional studies including those with patient data are required to fully understand their capabilities.

Fig. 1

Clinical PET/CT with CT contrast agent. The top row shows coronal (a) and transaxial (b) attenuation-corrected PET images. The arrows indicate a region that is believed to be an artifact from contrast agent. This conclusion is based upon an examination of the CT image showing probable contrast enhancement indicated by the arrow (c) and the uncorrected PET image in which abnormal tracer uptake is not evident.

Fig. 2

Schematic of the operation of the recovery algorithm. The PET data is first reconstructed using the CT-based attenuation correction. An iterative method consisting of two procedures is than applied. In the type 1 procedure, pixel-wise updates, within the dynamic region, of both attenuation and emission are performed using ICD[33]. In the type 2 procedure all emission values throughout the image are updated using OSEM[34].

Fig. 3

The true attenuation distribution for the chest (a) and shoulder (b) and the corresponding true activity distribution for the chest (c) and shoulder (d).

Fig. 4

Attenuation image with a 13-pixel artifact marked by arrow for the chest (a) and shoulder (e). The artifact was produced by assigning bone-valued attenuation coefficients to soft tissue. Sample emission reconstructions using the artifactual attenuation are also shown for varying sinogram count densities: 10⁶ (b), 10⁵ (c), and 10⁴ (d) for the chest slice. Parts (f–h) show similar reconstructions for the shoulder slice. In each case, OSEM reconstructions were performed using 4 iterations of 9 subsets with no post reconstruction filtering.

Fig. 5

Attenuation images similar to those of figure 4 containing two adjacent 13-pixel artifactual regions marked by an arrow for the chest (a) and the shoulder (b).

Fig. 6

Attenuation artifacts marked by arrow for the chest (a) and the shoulder (b) have the μ-value for soft tissue in place of bone.

Fig. 7

Attenuation images with single-pixel bone-valued artifacts in place of soft tissue are marked by arrows in the chest (a) and shoulder (b).

Fig. 8

Algorithm results for 100000 simulated sinogram counts: attenuation images. Top row shows the true attenuation map (left) and an attenuation map with an artifact marked by an arrow (right). The subsequent rows show the recovered attenuation map after each iteration of the algorithm described in the text.

Fig. 9

Algorithm results for 100000 simulated sinogram counts: activity images. The activity images in this figure correspond to the attenuation images in figure 9. Top row shows the activity images obtained with the true attenuation map (left) and the artifactual attenuation map (right). These images were reconstructed using OSEM with 9 subsets and 4 iterations. The subsequent rows show the emission image after each iteration of the algorithm described in the text.

Fig. 10

ROC curve for the task of detection of bone-like artifacts in the 100000 simulated count sinograms. P_tp and P_fp are true-positive and false-positive rates for artifact detection.

Fig. 11

2.1-cm² (13-pixel) bone-valued artifact in soft tissue in the chest (solid circles) and shoulder (open squares) - Plots of loglikelihood classification accuracy as a function of simulated sinogram count. The task was to identify the artifact-free attenuation image. Connecting lines are shown to guide the eye.

Fig. 12

2 2.1-cm² (13-pixel) artifacts in the chest (solid circles) and shoulder (open squares) - Plots of loglikelihood classification accuracy as a function of simulated sinogram count level. The task was to identify the image with one 13-pixel artifact compared to the image with two artifacts.

Fig. 13

2.1-cm² (13-pixel) soft-tissue-valued artifact in bone in the chest (solid circles) and shoulder (open squares)- Plots of loglikelihood classification accuracy as a function of simulated sinogram count level.

Fig. 14

Soft-tissue artifact with a μ-value intermediate between soft tissue and bone in the chest (solid circles) and shoulder (open squares) - Plots of accuracy as a function of simulated sinogram count level

Fig. 15

0.16-cm² (1-pixel) bone valued artifact in place of soft tissue in the chest (solid circles) and shoulder (open squares) - Plots of accuracy as a function of simulated sinogram count level.