Hi everyone,
I'm working on a semantic segmentation project in MATLAB involving medical images in DICOM format. My idea is use masks exported from Medical Image Labeler from Matlab. I used Medical Image Labeler to annotate several slices (around 10) of volumetric DICOM data. After exporting, I received a .mat file containing a groundTruthMedical object.
My goal is to train a segmentation network, and for that, I need to create a pixelLabelDatastore. I found two ways to do this:
1.Using a groundTruth object directly:
pxds = pixelLabelDatastore(gTruth);
…but this doesn’t work because my variable is of type groundTruthMedical, and MATLAB throws an error if I try to convert it directly to groundTruth due to incompatible DataSource types (e.g., VolumeSource).
groundTruthMedical with properties:
DataSource: [1×1 medical.labeler.loading.VolumeSource]
LabelDefinitions: [5×3 table]
2. Using the format:
pxds = pixelLabelDatastore(location, classNames, labelIDs);
…but in my case, the labels exist only inside the .mat file and not as individual PNG or image files saved on disk. I don't know what to provide for the location argument here.
Additionally, my original image set has over 600 DICOM slices, but only a few have been labeled, because I would like to see how does it work, as a smaller provisional test. I understand that combine(imds, pxds) requires exact correspondence between image and mask sets. So I probably need to create a subset of the imageDatastore to match the available masks — but I’m unsure of the best way to do this, especially when masks are still stored inside the groundTruthMedical object.My questions:
- What is the correct workflow to create a pixelLabelDatastore from a groundTruthMedical object?
- Is it possible to extract the labeled masks from groundTruthMedical and save them as PNGs to use with pixelLabelDatastore(location, classNames, labelIDs)?
- How do I properly subset my imageDatastore so that it matches only the labeled slices?
Any suggestions or examples would be greatly appreciated!
Thanks in advance.
