Implementation of the Australian Computer-Assisted Theragnostics (AusCAT) network for radiation oncology data extraction, reporting and distributed learning.

INTRODUCTION: There is significant potential to analyse and model routinely collected data for radiotherapy patients to provide evidence to support clinical decisions, particularly where clinical trials evidence is limited or non-existent. However, in practice there are administrative, ethical, technical, logistical and legislative barriers to having coordinated data analysis platforms across radiation oncology centres. METHODS: A distributed learning network of computer systems is presented, with software tools to extract and report on oncology data and to enable statistical model development. A distributed or federated learning approach keeps data in the local centre, but models are developed from the entire cohort. RESULTS: The feasibility of this approach is demonstrated across six Australian oncology centres, using routinely collected lung cancer data from oncology information systems. The infrastructure was used to validate and develop machine learning for model-based clinical decision support and for one centre to assess patient eligibility criteria for two major lung cancer radiotherapy clinical trials (RTOG-9410, RTOG-0617). External validation of a 2-year overall survival model for non-small cell lung cancer (NSCLC) gave an AUC of 0.65 and C-index of 0.62 across the network. For one centre, 65% of Stage III NSCLC patients did not meet eligibility criteria for either of the two practice-changing clinical trials, and these patients had poorer survival than eligible patients (10.6 m vs. 15.8 m, P = 0.024). CONCLUSION: Population-based studies on routine data are possible using a distributed learning approach. This has the potential for decision support models for patients for whom supporting clinical trial evidence is not applicable.

Modeling radiation pneumonitis of pulmonary stereotactic body radiotherapy: The impact of a local dose-effect relationship for lung perfusion loss.

PURPOSE: To investigate if a local dose-effect (LDE) relationship for perfusion loss improves the NTCP model fit for SBRT induced radiation pneumonitis (RP) compared to conventional LDEs. METHODS AND MATERIALS: Multi-institutional data of 1015 patients treated with SBRT were analyzed. Dose distributions were converted to NTD with α/ß = 3 Gy. The Lyman-Kutcher-Burman NTCP model was fitted to the incidence grade ≥2 RP by maximum likelihood estimation with mean lung dose (MLD), equivalent uniform doses (EUD) using three LDE functions (power-law (EUDpower), logistic with 2 free parameters (EUDlog-free) and logistic with fixed parameters describing local perfusion loss (EUDPerfusion)) and volume above a threshold dose (Vx). Models were compared with the Akaike weights (Aw) derived from the Akaike information criteria (AIC). RESULTS: The median time to grade ≥2 RP was 4.2 months and plateaued after 17 months at 5.4%. A strong dose-effect relationship for RP incidence was observed. The EUDPerfusion based NTCP model had the lowest AIC. The Aw were 0.53, 0.19, 0.11, 0.11, 0.05 for the EUDPerfusion, Vx, MLD, EUDlog-free and EUDpower LDEs respectively. CONCLUSION: A LDE for perfusion loss provided modest improvement in NTCP model fit for SBRT induced radiation pneumonitis.

Second malignant neoplasm risk after craniospinal irradiation in X-ray-based techniques compared to proton therapy.

Cranio-spinal irradiation (CSI) is widely used for treating medulloblastoma cases in children. Radiation-induced second malignancy is of grave concern; especially in children due to their long-life expectancy and higher radiosensitivity of tissues at young age. Several techniques can be employed for CSI including 3DCRT, IMRT, VMAT and tomotherapy. However, these techniques are associated with higher risk of second malignancy due to the physical characteristics of photon irradiation which deliver moderately higher doses to normal tissues. On the other hand, proton beam therapy delivers substantially lesser dose to normal tissues due to the sharp dose fall off beyond Bragg peak compared to photon therapy. The aim of this work is to quantify the relative decrease in the risk with proton therapy compared to other photon treatments for CSI. Ten anonymized patient DICOM datasets treated previously were selected for this study. 3DCRT, IMRT, VMAT, tomotherapy and proton therapy with pencil beam scanning (PBS) plans were generated. The prescription dose was 36 Gy in 20 fractions. PBS was chosen due to substantially lesser neutron dose compared to passive scattering. The age of the patients ranged from 3 to 12 with a median age of eight with six male and four female patients. Commonly used linear and a mechanistic doseresponse models (DRM) were used for the analyses. Dose-volume histograms (DVH) were calculated for critical structures to calculate organ equivalent doses (OED) to obtain excess absolute risk (EAR), life-time attributable risk (LAR) and other risk relevant parameters. A α' value of 0.018 Gy-1 and a repopulation factor R of 0.93 was used in the mechanistic model for carcinoma induction. Gender specific correction factor of 0.17 and - 0.17 for females and males were used for the EAR calculation. The relative integral dose of all critical structures averaged were 6.3, 4.8, 4.5 and 4.7 times higher in 3DCRT, IMRT, VMAT and tomotherapy respectively compared to proton therapy. The mean relative LAR calculated from the mean EAR of all organs with linear DRM were 4.0, 2.9, 2.9, 2.7 higher for male and 4.0, 2.9, 2.8 and 2.7 times higher for female patients compared to proton therapy. The same values with the mechanistic model were 2.2, 3.6, 3.2, 3.8 and 2.2, 3.5, 3.2, 3.8 times higher compared to proton therapy for male and female patients respectively. All critical structures except lungs and kidneys considered in this study had a substantially lower OED in proton plans. Risk of radiation-induced second malignancy in Proton PBS compared to conventional photon treatments were up to three and four times lesser for male and female patients respectively with the linear DRM. Using the mechanistic DRM these were up to two and three times lesser in proton plans for male and female patients respectively.

Estimating Second Malignancy Risk in Intensity-Modulated Radiotherapy and Volumetric-Modulated Arc Therapy using a Mechanistic Radiobiological Model in Radiotherapy for Carcinoma of Left Breast.

OBJECTIVES: The aim of this study is to estimate second cancer risk (SCR) in intensity-modulated radiotherapy (IMRT) and volumetric-modulated arc therapy (VMAT) using a mechanistic radiobiological model. The model also takes into account patient age at exposure and the gender-specific correction factors of SCR. MATERIALS AND METHODS: Fifty IMRT and VMAT plans were selected for the study. Monte Carlo-based dose calculation engine was used for dose calculation. Appropriate model parameters were taken from the literature for the mechanistic model to calculate excess absolute risk (EAR), lifetime attributable risk, integral dose and relative risk (RR) for lungs, contralateral breast, heart, and spinal cord. RESULTS: The mean monitor unit (MU) in IMRT and VMAT plans were 751.1 ± 133.3 and 1004.8 ± 180, respectively, for IMRT and VMAT. The mean EAR values with age correction were 44.6 ± 11.9, 11.2 ± 6.4, 5.4 ± 4.0, 1.4 ± 0.5, and 0.3 ± 0.2 for left lung, right lung, contralateral breast, heart, and spinal cord, respectively, for the IMRT treatments and 54.6 ± 20.6, 30.2 ± 12.0, 13.8 ± 8.6, 1.6 ± 0.6, and 0.9 ± 0.5 for the VMAT treatments in units of 10,000 PY. The RR of 6.7% and 9.1%, respectively, for IMRT and VMAT found in our study using computational models is in close comparison with the value reported in a large epidemiological breast cancer study. CONCLUSIONS: VMAT plans had a higher risk of developing second malignancy in lung, contralateral breast, heart, and cord compared to IMRT plans. However, the increase in risk was found to be marginal compared to IMRT. Incorporating the age correction factor decreased the risk of contralateral breast SCR. No strong correlation was found between EAR and MU.

Impact of microscopic disease extension, extra-CTV tumour islets, incidental dose and dose conformity on tumour control probability.

The impact of microscopic disease extension (MDE), extra-CTV tumour islets (TIs), incidental dose and dose conformity on tumour control probability (TCP) is analyzed using insilico simulations in this study. MDE in the region in between GTV and CTV is simulated inclusive of geometric uncertainties (GE) using spherical targets and spherical dose distribution. To study the effect of incidental dose on TIs and the effect of dose-response curve (DRC) on tumour control, islets were randomly distributed and TCP was calculated for various dose levels by rescaling the dose. Further, the impact of dose conformity on required PTV margins is also studied. The required PTV margins are ~2 mm lesser than assuming a uniform clonogen density if an exponential clonogen density fall off in the GTV-CTV is assumed. However, margins are almost equal if GE is higher in both cases. This shows that GE has a profound impact on margins. The effect of TIs showed a bi-phasic relation with increasing dose, indicating that patients with islets not in the beam paths do not benefit from dose escalation. Increasing dose conformity is also found to have considerable effect on TCP loss especially for larger GE. Further, smaller margins in IGRT should be used with caution where uncertainty in CTV definition is of concern.

A method to obtain correct standard uptake values in Pinnacle treatment planning system for target volume delineation.

Standardized uptake value (SUV) is an advanced tool for quantitative tumor identification and metabolic target volume delineation (TVD) in diagnostic and therapeutic settings. It is thus important to establish a quality assured process to maintain the traceability of data correctly by positron emission tomography (PET) systems. Patient administration of 18fluoro-deoxy-glucose is increasingly delivered by automated infusion systems (AISs). Whenever AIS is used, its accuracy and traceability measurement need verification. In addition, it was observed that the unreproducible SUV displayed in PET and the treatment planning system (TPS) may cause grave concerns for radiation oncologists for TVD. This concern may complicate the correlation of TVD on PET and TPS and their clinical reporting. The SUV traceability was established from the PET system to AIS. Its accuracy was verified by cross-referencing to the reference dose calibrator traceable to a primary standard. The SUV values were converted in TPS using the in-house "clinical tool" to be identical as in PET, to allow radiation oncologists to use SUV confidently. The outcome of this study enables the clinical groups to rely on the correct SUV values displayed on the TPS and to improve the quality of care for patients in clinical procedures.

Loss of local control due to tumor displacement as a function of margin size, dose-response slope, and number of fractions.

PURPOSE: Geometric uncertainties are inevitable in radiotherapy. To account for these uncertainties, a margin is added to the clinical target volume (CTV) to create the planning target volume (PTV), and its size is critical for obtaining an optimal treatment plan. Dose-based (i.e., physical) margin recipes have been published and widely used, but it is important to consider fractionation and the radiobiological characteristics of the tumor when deriving margins. Hence a tumor control probability (TCP)-based margin is arguably more appropriate. METHODS: Margins required for ≤ 1% loss in mean population TCP (relative to a static tumor) for varying numbers of fractions, varying slope of the dose-response curve (γ50) and varying degrees of dose distribution conformity are investigated for spherical and four-field (4F)-brick dose distributions. To simulate geometric uncertainties, systematic (Σ) and random (σ) tumor displacements were sampled from Gaussian distributions and applied to each fraction for a spherical CTV. Interfraction tumor motion was simulated and the dose accumulated from fraction to fraction on a voxel-by-voxel basis to calculate TCP. PTV margins derived from this work for various fraction numbers and dose-response slopes (γ50) for different degrees of geometric uncertainties are compared with margins calculated using published physical-dose- and TCP-based recipes. RESULTS: Larger margins are required for a decrease in the number of fractions and for an increase in γ50 for both spherical and 4F-brick dose distributions. However, the margins can be close to zero for the 4F-brick distribution for small geometric uncertainties (Σ = 1, σ = 1 mm) irrespective of the number of fractions and the magnitude of γ50 due to the higher "incidental" dose outside the tumor. For Σ = 1 mm and σ = 3 mm, physical-dose-based recipes underestimate the margin only for the combination of hypofractionated treatments and tumors with a high γ50. For all other situations TCP-based margins are smaller than physical-dose-based recipes. CONCLUSIONS: Margins depend on the number of fractions and γ50 in addition to Σ and σ. Dose conformity should also be considered since the required margin increases with increasing dose conformity. Ideally margins should be anisotropic and individualized, taking into account γ50, number of fractions, and the dose distribution, as well as estimates of Σ and σ. No single "recipe" can adequately account for all these variables.