When I am trying to train a Deep Learning model using DICOM files in Python, I first extract the pixel array of the DICOM files using Pydicom and then input the array as NumPy arrays to the network. If I want to be able to measure quantitative metrics like SUV of the model outputs (PET images) which are NumPy arrays, how should I proceed? If the answer is to save as DICOM files, how should that be done to be able to compare output and input DICOM files?
"If the answer is to save as DICOM files" (which is not quite clear from your question), then you could again use pydicom to save them. You could use the original dicom files (loaded by pydicom) as template and just modify/delete those dicom tags that do not fit. I.e., use a template dicom file "dcm" and replace dcm.PixelData by your new slice data (e.g., dcm.PixelData = newdata.astype(np.uint16)). Add some new and unique UIDs (see, e.g., https://www.medicalconnections.co.uk/FreeUID) and then use dcm.save_as().
But probably there is a better format than DICOM to save and process your new data.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts