Work Description

Title: Virtual patient phantom dataset for DblurDoseNet
Author: Zongyu Li

This dataset provides training and testing phantoms as well
as DVKs that are used in paper "DblurDoseNet: A Deep Residual Learning Network
for Voxel Radionuclide Dosimetry Compensating for SPECT Imaging Resolution"
by Zongyu Li, Jeffrey A. Fessler, Justin K. Mikell, Scott J. Wilderman
and Yuni K. Dewaraja. Accepted by Medical Physics, 2021. DOI: 10.1002/mp.15397.

Motivation: Current methods for patient-specific voxel-level dosimetry in radionuclide therapy suffer from a trade-off
between accuracy and computational efficiency. Monte Carlo (MC) radiation transport algorithms are considered the gold standard
for voxel-level dosimetry but can be computationally expensive, whereas faster dose voxel kernel (DVK) convolution
can be sub-optimal in the presence of tissue heterogeneities. Furthermore, the accuracies of both these methods are limited
by the spatial resolution of the reconstructed emission image. To overcome these limitations,
we considers a single deep convolutional neural network (CNN) with residual learning (named DblurDoseNet)
that learns to produce dose-rate maps while compensating for the limited resolution of SPECT images.
We took the novel approach of constructing a convolutional neural network with residual learning
to handle the accuracy-efficiency tradeoff while compensating for the limited resolution of SPECT images.
We then test our CNN on clinically relevant phantoms and patients undergoing Lu-177 DOTATATE therapy in our clinic.
Our network demonstrated superior results than Monte Carlo, the current gold standard for voxel dosimetry,
but only takes a fraction of time. Thus, the DblurDoseNet has the potential for real-time patient-specific dosimetry
in clinical treatment planning due to its demonstrated improvement in accuracy, resolution,
noise and speed over the DVK/MC approaches. Matlab is needed to access the phantoms and Python
(with Numpy package installed) is needed to access the DVKs.

To generate SPECT activity maps, we first collected 14 Ga-68 DOTATATE PET/CT images from our Siemens mCT scanner
and reconstructed using the standard clinic protocol: 3D ordered subset expectation maximization (OSEM) with 3 iterations,
21 subsets that included resolution recovery, time-of-flight (TOF), and a 5mm (FWHM) Gaussian post-reconstruction filter.
Next, we ran SIMIND MC code to generate Lu-177 SPECT projections using the PET images as activity maps.
Finally, we used an in-house 3D OSEM algorithm with CT-based attenuation correction, triple energy window scatter correction
and collimator-detector response modeling (4 subsets and 16 iterations, 128×128×81 matrix with voxel size 4.8×4.8×4.8mm^3,
no Gaussian smoothing) to reconstruct the SPECT images. Density maps were generated using an experimentally derived
CT-to-density calibration curve from Ga-68 DOTATATE PET/CT images. Dose voxel kernel(DVK) dose-rate maps are the convolution
between SPECT reconstruction and DVKs (considering both beta particles and gamma rays).
Ground truth (GT) dose-rate maps were generated by running 1 billion histories of DPM MC simulation
inputting with the true activity map and the density map. Monte Carlo (MC) dose-rate maps were generated with the same procedure
but inputting with the SPECT reconstruction images instead of true activity map.
All images were in ".mat" format, having size 512x512x130 with voxel size 0.98x0.98x3mm^3.

The data is organized as follows:
The virtual patient phantom data is splitted into two folders,
namely `train` and `test` that denote training data and testing data,
respectively.

The training data contains the following phantoms:
-`petvp01`
-`petvp04`
-`petvp11`
-`petvp15`
-`petvp17`
-`petvp19`
-`petvp21`
-`petvp22`
-`petvp23`
-`petvp24`

The testing data contains the following phantoms:
-`petvp6`
-`petvp7`
-`petvp8`
-`petvp10`
-`petvp16`
-`petvp25`

In each phantom (folder), the following ".mat" files are provided:
-`xtrue`: the true activity map, corresponding to the true activity defined in Ga-68 PET images.
-`spect`: the SPECT reconstruction image, by running 16 iterations, 4 subset OSEM algorithm.
-`density`: the density map, in (x1000) g/cm^3.
-`doseGT`: the ground-truth dose-rate map, by running 1 billion histories of DPM with the true activity map.
-`doseMC`: the MC dose-rate map, by running 1 billion histories of DPM with the SPECT reconstruction image.
-`doseDVK`: the DVK dose-rate map, by convolving the SPECT reconstruction image with DVKs.
-`mask`: the whole-body mask.

In addition to the phantom data, the DVKs are also provided (in "DVKs" folder),
these kernels are in ".npy" format (and ".mat" format) and can be loaded using Python or Matlab:
-`beta.npy`: the beta kernel (of size 9x9x9) of Lu-177 in tissues.
-`gamma.npy`: the gamma kernel (of size 99x99x99) of Lu-177 in tissues.