Segmentation on 3D DICOM ultrasound fetus image


I am new to VTK (and ITK) and I succesfully visualized with volume rendering my 3D dataset. However, I now only want to show the fetus and the placental volume. So I want to segment the fetus and the placenta, and remove the noise and abdominal.

I already tried with VolView (and 3D slicer) to create a mask and get a hands on experience before generating my python code, but I dont get a proper mask (either automated or semi-automated with seed points). There are a lot of examples for CT and MR, but can someone help me segmenting an ultrasound image?

Thank you

Hi, Romy

From the programming standing point, there is no difference between CT, MR and ultrasound images. They’re all data sets with floating pointing numbers.

Now, to your problem. In petroleum industry we have something called a seismic survey, which is much like a big ultrasound image of the Earth :slightly_smiling_face:. I mean, they’re both images of sound wave reflections caused by contrasts of sound impedances and not direct images of hard and soft rocks or hard and soft tissues in your case. This is your challenge.

To get direct images you have to do some processing prior to segmenting it or using an advanced segmentation method (possibly with the help of machine learning techniques). In petroleum industry we use a class of complex algorithms called seismic inversion (that is, given the reflections, give me the rock types). So, I guess US images likewise require some kind of sophisticated processing, which is far beyond the scope of this forum. I’d recommend reading the papers suggested here: . You can also search Google Scholar for the most recent research ( >= 2019 ) in the subject: . I hope this helps you set in the right track.

all the best,


1 Like