Dec 1
1- Tests on initialisation using the parameters that have been recorded (angles and centre of rotation).
2- Initialisation works well on images acquired on the same scanner.
3- Alternatives:
3a- identification of the object of interest
3b- Registration
4- Conclusion of the meeting: Registration will be tested.
Dec 8
1- Initialisation through registration works fine in all but one case (this case usually fails).
1a- Affine and rigid were tested. I personally prefer the rigid registration approach as it does not change the size of the original heart.
2- Registration is very slow.
3- Future work: Assure robustness.
Dec 14 – Segmentation and detection meeting
Attendants: Jan Deprest (JD), Wenqi Li (WL), Tom Vercauteren (TV), Guotai Wang (GW), Maria A Zuluaga (MAZ)
1- Definition of targeted applications:
a. Adding degrees of freedom on the position of the ultrasound transducer during open spina bifida surgery.
b. Heart monitoring during in utero cardiac interventions.
For this, we will target heart detection (WL) followed by standard view obtention (MAZ).
2- Data will be 3D US from 22-26 weeks foetus. Typically, it will have a 4 chamber view of the heart and a slightly randomly oriented view of the fetal thorax. The goal is to replicate a plausible orientation from the open spina bifida surgery protocol.
3- JD will contact people in Leuven as to provide us with 50 cases.
4- In parallel, JD will try to gather CHD data (more rare). The problem here will be location of stomach, liver and lungs.
5- WD and MAZ will contact Roz to get an echo acquisition of the phantom.
6- Validation for GW:
a. Slic-Seg should be assessed on abnormal placentals and through confrontation with Michael A.
b. For the latter, GW should contact Michael and arrange a meeting (virtual or real). To note, that GW needs to reapply for a visa if he needs to come to Leuven.
c. For evaluation with abnormal placentas we will require additional datasets from accreta and TTTS.
Dec 21 – Segmentation and detection meeting
Attendants: Jan Deprest (JD), Wenqi Li (WL), Guotai Wang (GW), Maria A Zuluaga (MAZ)
1- JD will come on Jan 18 with two more people to discuss about the project: Isabelle, a gynaecologist specialist in twins, and Michael A., the radiologist.
2- We will all meet again on Jan 11 to prepare for this meeting.
3- During the meeting on the 18th, each of us will give a brief presentation (5 mins) to show to the clinicians what is your project about and what we think we need from them.
4- JD briefly described the types of images that are used at each step of the clinical workflow/pipeline:
a. Pre-op images: 2D US, 3D US (not commonly used. Will be for us and has the issue that is protected by GE) and MRI.
b. Op images: 2D US (VGA sequence. Not timestamped at the moment), fetoscope, motion tracker, maternal landmarks.
c. Post-op: 2D US and MRI.
5- Important landmarks to be identified from MRI: Placenta, Cord, lungs, liver, heart, stomach.
6- JD described the procedure used by clinicians to identify important landmarks. Of relevance is the concept of best available plane (see Nawapun et al, 2015) and how this is achieved. Doing this automatically could be a project in itself.
a. WL will work on landmark detection in fetal MR (existing data). We’ll switch to 3D US once data arrived.
b. JD mentioned in fetal MR images — signals (voxel intensities/texture) of the fetus brain, lungs, bladder, and spine fluid are very similar.
7- JD mentioned to GW that the segmentation of the foetal body is of relevance as it can be used clinically to measure the foetus volume.
8- JD will perform a foetal heart acquisition on Jan 11 using the phantom.