Search

Additional file 1. Figure S1. Schematic illustration of the u-shaped convolutional neural network employed in this study (example 64 → 64 projections). Figure S2. Comparison of measured and simulated projections of a NEMA phantom with synthetic generated projections by u-nets U1 and U4. The right graphs each show the cross section through the projections along the colored lines, where the solid curves represent the simulations and the dashed curves represent the measurements. The noise-free projection (simulation) is also shown for both the measured and simulated projections (gray curve). Figure S3. Training and validation loss curves for the training of u-nets U1 and U3. Since a different number of training epochs and data were selected for the training of both networks, the relative training step is given on the x-axis. Therefore, a training step of 100% corresponds to 60 epochs for u-net U1, while it corresponds to 200 epochs for u-net U2.

Data and documentation from the EMPIR MRTDosimetry (15HLT06 ) and PINICAL_MRT (19SIP01) projects

Background Differentiated thyroid cancer has been treated with radioiodine for almost 80 years, although controversial questions regarding radiation-related risks and the optimisation of treatment regimens remain unresolved. Multi-centre clinical studies are required to ensure recruitment of sufficient patients to achieve the statistical significance required to address these issues. Optimisation and standardisation of data acquisition and processing are necessary to ensure quantitative imaging and patient-specific dosimetry. Material and methods A European network of centres able to perform standardised quantitative imaging of radioiodine therapy of thyroid cancer patients was set-up within the EU consortium MEDIRAD. This network will support a concurrent series of clinical studies to determine accurately absorbed doses for thyroid cancer patients treated with radioiodine. Five SPECT(/CT) systems at four European centres were characterised with respect to their system volume sensitivity, recovery coefficients and dead time. Results System volume sensitivities of the Siemens Intevo systems (crystal thickness 3/8″) ranged from 62.1 to 73.5 cps/MBq. For a GE Discovery 670 (crystal thickness 5/8″) a system volume sensitivity of 92.2 cps/MBq was measured. Recovery coefficients measured on three Siemens Intevo systems show good agreement. For volumes larger than 10 ml, the maximum observed difference between recovery coefficients was found to be ± 0.02. Furthermore, dead-time coefficients measured on two Siemens Intevo systems agreed well with previously published dead-time values. Conclusions Results presented here provide additional support for the proposal to use global calibration parameters for cameras of the same make and model. This could potentially facilitate the extension of the imaging network for further dosimetry-based studies.