Hi,
I am just new to this project. I am just wondering if the dataset given has already been standardized or normalized so that the pixel intensity values for pet are so small?
Hi,
No, the images were not standardized nor pre-processed. They are stored in Hounsfield Unit (HU) and Standardized Uptake Value (SUV) for CT and PET, respectively.
The CT images have a dynamic range of approximately [-1000, 2000] HU and the PET images of [0, 100] SUV. Usually, in PET images, a region with a SUV above 2.5 can be considered suspicious, but, this value can vary a lot and it is not uncommon to find tumors with SUV above 10 (and even, in some cases, above 100).
It is important to keep in mind that SUV is a semi-quantitative unit, thus standardizing the PET images is crucial.
Let me know if you need more clarification.
Best,
Valentin
Hi Voreille,
Thank you so much for your thorough reply. May I bore you with a small question. When I segmented another set of PET images for lung tumors before (it was stored in Dicom format), the pixel intensities for the tumor part usually reach 20000 and above. Does this mean that the images I was segmented were not stored as SUV and the intensity values need to be converted? Is this standardization a suggested way to make? I am looking forward to your reply.
Best,
Wenhui
Hi,
Depending on the manufacturer, the PET Dicom could be stored as counts, and it is in the range you describe. The SUV is the unit used by clinicians for tumor follow-up as it is more or less reproducible on different acquisitions for the same patient, the same scanner, and the same protocol. In the context of image analysis, I would not trust too much this unit since it can vary quite a lot between patients and scanners. For that reason, it is important to normalize (or standardize) your images, thus, it should not be an issue if the units are not the same as long as the normalization (or standardization) removes this dependency.
Best,
Valentin