Image Registration Uncertainty in Head and Neck Cancer

Factors affecting MRI-CT fusion accuracy include: 1) Patient positioning differences: the neck was extended on CT but neutral on MRI, causing significant anatomical differences, especially in the neck and jaw. 2) Weight change (2 kg) can alter soft tissue contours. 3) Different acquisition days: tumor may have changed, and bowel/bladder filling may differ (though less relevant for head and neck). 4) MRI geometric distortion, especially at 3 T due to susceptibility and gradient nonlinearity, which can cause spatial inaccuracies. 5) Differences in slice thickness and resolution. 6) The use of different coils and immobilization devices; the MRI may not have used the same mask, so the head position may differ. 7) The registration algorithm (rigid vs. deformable) and its parameters. To evaluate registration, perform visual inspection: check alignment of bony landmarks (e.g., skull base, cervical spine, mandible) and soft tissue structures (e.g., carotid arteries, muscles). Use a checkerboard or blend image. Assess the distance between corresponding points. If misalignment is significant, consider re-registration with a different algorithm (e.g., deformable) or repeat imaging with consistent positioning. Also, quantify using mutual information or other metrics, but visual check is key. For head and neck, rigid registration may be insufficient due to neck flexion; a deformable registration might be needed but can introduce errors. Ultimately, the radiation oncologist must review and approve the fusion.

To optimize registration: 1) Use a rigid registration first, focusing on the skull base and upper cervical spine, which are relatively fixed. 2) Then apply a deformable registration (e.g., demons or B-spline) to account for neck flexion differences, but carefully review the deformation in the tumor region. 3) Mask the region of interest (e.g., exclude the shoulders and mandible if they are not relevant) to reduce the influence of positioning differences. 4) Use the contrast-enhanced MRI for better soft tissue contrast. 5) Ensure the MRI and CT are both acquired with the same immobilization mask if possible; if not, consider repeating the MRI with the mask. 6) Use a registration algorithm that is robust to intensity differences (e.g., mutual information). Limitations: deformable registration can introduce false deformations, especially in areas of poor contrast or near metal artifacts. It may not be able to correct for large positional differences. The registration is an approximation; the radiation oncologist must verify the target delineation on the fused images and may need to adjust. If the fusion is inadequate, consider repeating the MRI with proper positioning or using CT alone with clinical judgment.

If fusion is inaccurate in the base of tongue, alternatives include: 1) Repeat the MRI with the patient in the same position as CT (using the mask and neck extension). This is the best solution if feasible. 2) Use the CT alone with clinical examination and endoscopic findings to delineate the tumor, accepting larger margins. 3) Use PET/CT if available, as FDG-PET can help localize the tumor. 4) Use ultrasound for tongue base? Not practical. 5) Consider a second MRI with a different protocol (e.g., faster sequences) to reduce motion. 6) Use intraoperative findings if surgery is planned. 7) Consult with a neuroradiologist for direct review of the MRI and CT side-by-side. The decision should involve the radiation oncologist and may require a multidisciplinary discussion. If the tumor is not well visualized, the radiation oncologist may choose to treat with larger margins or use a boost technique. Ultimately, patient safety and tumor control are paramount.