Organ-wise mean Standardized Uptake Value (SUV) Calculation Framework based on PET/CT Images

Framework to calculate mean organ Standardized Uptake Values (SUVs) from PET/CT images. The CT-based organ segmentations are performed by TotalSegmentator.

Getting Started:

Input:

• Path to a folder where for each individual patient a folder is present. These folder names will end up as IDs in the output table.
• Each patient folder should contain a folder with a DICOM series for both the PET and according CT image:
  • A PET folder with a name containing "PET", "FDG", or "5-MIN"
  • A CT folder with a name containing "CT"
• Dictionary with organs of interest. Defined in the SUV_Extractor.py file.
• A list of organs of interest

Pipeline Configuration Flags:

• SEGMENTATE_ORGANS: If True, performs automatic segmentation of individual organs from CT images.
• SEGMENTATE_BODY: If True, performs automatic segmentation of a whole-body mask from CT images.
• SEGMENTATE_LUNG: If True, If True, performs dedicated lung segmentation using the "lung_nodules" task in the TotalSegmentator.
• NORMALIZE: If True, normalizes PET intensities using an aorta-based reference (blood pool normalization).
• ERODE: If True, applies morphological erosion to organ masks before extracting PET values.

Output:

• Table with a row per patient and the according mean SUV values for the organs of the patient.

Calculation of SUVs:

The Script uses the following general formula to calculate for each voxel the SUVs:

$$\mathrm{SUV}_{bw}(x,y,z)=\frac{C(x,y,z)\cdot W}{D_{\mathrm{corr}}}$$

where

• $C(x,y,z)$ is the voxel activity concentration in Bq/mL.
• $W$ is the patient body weight in g
• $D_{\mathrm{corr}}$ is the decay-corrected injected Dose in Bq.

Decay Correction

The $D_{\mathrm{corr}}$ is obtained with the exponential decay function

$$D_{\mathrm{corr}} = D_0 \, e^{-(\ln 2 / T_{1/2})\Delta t}$$

This equation can then be plugged into the SUV equation. Now we just need to consider two possible cases depending on the PET data.

BQML Case:

If the original PET image is already provided as an Bq/mL map the units do not have to be converted. Here the intensity map is already the Activity Concentration.

$$C(x,y,z)=I(x,y,z)$$

So the following formula is used:

$$\mathrm{SUV}_{bw}(x,y,z)=\frac{I(x,y,z)\cdot W}{D_0 \, e^{-(\ln 2 / T_{1/2})\Delta t}}$$

Non BQML Case:

If the orignal PET image is not provided as a Bq/mL map we need to convert it. This is done using a raw PET conversion with

$$C(x,y,z)=I(x,y,z)\cdot s+b$$

where

• $I(x,y,z)$ is the raw voxel intensity value
• $s$ is the RescaleSlope
• $b$ is the RescaleIntercept

Using this formula we end up with the following final formula:

$$\mathrm{SUV}_{bw}(x,y,z)=\frac{(I(x,y,z)\cdot s+b)\cdot W}{D_0 \, e^{-(\ln 2 / T_{1/2})\Delta t}}$$