Inaccurate Reference Image Selection Causes Registration Artifacts in Multi-Tracer Alignment

[LalithShiyam]

Affected Component:

Image registration module (alignment of multiple PET/CT tracer images).

Issue Description:

The current implementation selects a reference image at random for aligning multiple PET/CT tracer images (e.g., FDG, DOPA, Fluciclovine) of the same patient. The assumption is that the reference image should ideally have the smallest field-of-view (FOV) for successful alignment. However, the current selection algorithm, which calculates the total volume to determine the smallest FOV, is not robust and occasionally selects an unsuitable reference image. This leads to significant registration artifacts in the aligned images.

Steps to Reproduce:

1. Run pumaz with a subject having disparate field-of-view.
2. And observe the aligned images, you can clearly see the artifacts.

Expected Results:

The reference image with the smallest FOV should be accurately selected, resulting in well-aligned images without registration artifacts.

Actual Results:

The algorithm sometimes selects an inappropriate reference image, leading to poor alignment and noticeable registration artifacts.

Suggested Fix:

A more robust algorithm needs to be developed for selecting the reference image based on field-of-view criteria, possibly incorporating additional metrics for reliability.

Severity:

High – this affects the primary functionality of the tool and may compromise the quality of the medical imaging data.

[LalithShiyam]

This issue is solved in the commit here!