Hello! I have a series of 2D (untracked) echo data from several patients, which contains both sweeps and still shots with ECG on the bottom of the frame. I would like to obtain 3D models from this data, but I’m unsure how to even get started. I’ve been able to upload all of the DICOM files of the data, and was told that I need “ground truth” data, but I’m not entirely sure what that entails. Any advice is appreciated!
Hi Madisyn!
First, on the ECG. Do you need to gate the ultrasound sweeps based on the ECG data? Maybe you have cardiac data or vessel data?
So, independent of the ECG, you will need to add a tracking for the 3D pose information. You can run the algorithm “Convert to Sweep” which will expect a tracking sequence file as input. I would recommend you to use a csv file for that. Each row corresponds to one pose. It can be expressed either with euler angles or a 4x4 flattened pose matrix (first column flattened followed by the second column flattened etc). Additionally, a timestamp is expected afterwards and a quality value where 1 is perfect quality.
If the Convert to Sweep algorithm does not show up, check whether the modality of the object is set to Ultrasound. This can be done by right-clicking on the data > set modality and choosing Ultrasound.
If the sweep you see subsequently is flat in the 3D seemingly wrong, check whether your sweep data has timestampts. If it does not, the tracking file should have poses for each frame and not more than that.
After you have a valid sweep that can also be viewed in the MPR and 3D views, you can also run the Sweep Compounding algorithm to produce a 3D volume if this is the data you require.
As for the ground truth data you are referring to - I am not sure what that would necessarily be. What is the problem you are trying to solve?
Hello! Unfortunately, I do not have any kind of tracking data or a tracking sequence file. The machines that recorded the ultrasound data did not capture any positional data for the scans. I think this is why ground truth data was mentioned, although I am unsure.
The 3D position of the scans needs to be set to work with the ultrasound in 3D. This can be made up (fake) or you can try to predict it with a model of course e.g GitHub - ImFusionGmbH/DualTrack: DualTrack: Sensorless 3D Ultrasound needs Local and Global Context (pay attention that this is trained and validated on a particular dataset that might not correspond to you scan).
May I ask again, what is the problem that you aim to solve?
Hm okay, good to know! I’ll try to look into that more. I’m trying to obtain a series of 3D models from 2D echo data across the cardiac cycle so that I might use them for CFD. The specific anatomy I am working with is pre-operational TGA.