Clarification Request on Task1 Challenge

Dear SynthRAD_2023 Community,

I trust this message finds you well. Firstly, I would like to express my sincere appreciation to the entire team of specialists who diligently contributed to the creation and maintenance of this valuable dataset.

My query pertains to a particular challenge presented in Task 1. I have noticed an inconsistency when observing MRI images in which the facial region appears to be removed. However, this doesn't seem to align with the same region in corresponding CT images. For your reference, I am specifically referring to the images in the masked region 1BA032_z11.

As per my current understanding, our objective is to generate synthetic CT (sCT) images from the given MRI inputs, and subsequently compare these with the standard CT images. If this understanding is incorrect, I kindly request your guidance and clarification.

In the case that my interpretation is correct, I'm curious if there exists any methodology or approach we might employ to circumnavigate this apparent discrepancy between the MRI and CT images.

I appreciate your time and consideration in addressing my queries. I am certain that your guidance will significantly contribute to my understanding and performance in this challenge.

Thank you in advance for your assistance.

Kind Regards, Convez376

Dear Convez76,

Thank you for daring to reach out! We highly appreciate your message.

You have correctly spotted that the filed-of-view (FOV) extension is not necessarily the same on the different modalities, given that the two acquisitions are independent. Your interpretation is correct: the goal is to generate sCT based on the input image, so in this case the MRI.

We are fully aware of possible mismatches in the body contour, or remaining mismatches after rigid registration. To partially take into account this, we will make use of a mask based on the intersection of the two body contours, i.e., MRI and CT. Especially at the cranio-caudal extremities of the FOV, the mismatch may be large, as for the case you pointed out.

What to do with this? Of course, this is up to you, and the approach you have decided to consider. However, in the case of a supervised one, where pairing/matching is most likely critical, one could consider applying for additional registration, e.g., deformable, or removing the incriminated slices.

We think that this is precisely where it gets interesting to compare what different participants will decide to do. So, I am looking forward to see which path you will take.

I hope my answer addresses your question, and if not, please do not hesitate to reach out once more.

Take care, and good luck with the preparation for your submission!

Matteo