Prescribing and evaluating target dose in dose-painting treatment plans

Prediction of radiologic outcome-optimized dose plans and post-treatment magnetic resonance images: A proof-of-concept study in breast cancer brain metastases treated with stereotactic radiosurgery

Background and purpose

Information in multiparametric Magnetic Resonance (mpMR) images is relatable to voxel-level tumor response to Radiation Treatment (RT). We have investigated a deep learning framework to predict (i) post-treatment mpMR images from pre-treatment mpMR images and the dose map ("forward models"), and, (ii) the RT dose map that will produce prescribed changes within the Gross Tumor Volume (GTV) on post-treatment mpMR images ("inverse model"), in Breast Cancer Metastases to the Brain (BCMB) treated with Stereotactic Radiosurgery (SRS).

Materials and methods

Local outcomes, planning computed tomography (CT) images, dose maps, and pre-treatment and post-treatment Apparent Diffusion Coefficient of water (ADC) maps, T1-weighted unenhanced (T1w) and contrast-enhanced (T1wCE), T2-weighted (T2w) and Fluid-Attenuated Inversion Recovery (FLAIR) mpMR images were curated from 39 BCMB patients. mpMR images were co-registered to the planning CT and intensity-calibrated. A 2D pix2pix architecture was used to train 5 forward models (ADC, T2w, FLAIR, T1w, T1wCE) and 1 inverse model on 1940 slices from 18 BCMB patients, and tested on 437 slices from another 9 BCMB patients.

Results

Root Mean Square Percent Error (RMSPE) within the GTV between predicted and ground-truth post-RT images for the 5 forward models, in 136 test slices containing GTV, were (mean ± SD) 0.12 ± 0.044 (ADC), 0.14 ± 0.066 (T2w), 0.08 ± 0.038 (T1w), 0.13 ± 0.058 (T1wCE), and 0.09 ± 0.056 (FLAIR). RMSPE within the GTV on the same 136 test slices, between the predicted and ground-truth dose maps, was 0.37 ± 0.20 for the inverse model.

Conclusions

A deep learning-based approach for radiologic outcome-optimized dose planning in SRS of BCMB has been demonstrated.

The impact of organ motion and the appliance of mitigation strategies on the effectiveness of hypoxia-guided proton therapy for non-small cell lung cancer

BACKGROUND AND PURPOSE

To investigate the impact of organ motion on hypoxia-guided proton therapy treatments for non-small cell lung cancer (NSCLC) patients.

MATERIALS AND METHODS

Hypoxia PET and 4D imaging data of six NSCLC patients were used to simulate hypoxia-guided proton therapy with different motion mitigation strategies including rescanning, breath-hold, respiratory gating and tumour tracking. Motion-induced dose degradation was estimated for treatment plans with dose painting of hypoxic tumour sub-volumes at escalated dose levels. Tumour control probability (TCP) and dosimetry indices were assessed to weigh the clinical benefit of dose escalation and motion mitigation. In addition, the difference in normal tissue complication probability (NTCP) between escalated proton and photon VMAT treatments has been assessed.

RESULTS

Motion-induced dose degradation was found for target coverage (CTV V_95% up to -4%) and quality of the dose-escalation-by-contour (Q_RMS up to 6%) as a function of motion amplitude and amount of dose escalation. The TCP benefit coming from dose escalation (+4-13%) outweighs the motion-induced losses (<2%). Significant average NTCP reductions of dose-escalated proton plans were found for lungs (-14%), oesophagus (-10%) and heart (-16%) compared to conventional VMAT plans. The best plan dosimetry was obtained with breath hold and respiratory gating with rescanning.

CONCLUSION

NSCLC affected by hypoxia appears to be a prime target for proton therapy which, by dose-escalation, allows to mitigate hypoxia-induced radio-resistance despite the sensitivity to organ motion. Furthermore, substantial reduction in normal tissue toxicity can be expected compared to conventional VMAT. Accessibility and standardization of hypoxia imaging and clinical trials are necessary to confirm these findings in a clinical setting.

Combined quality and dose-volume histograms for assessing the predictive value of 99mTc-MAA SPECT/CT simulation for personalizing radioembolization treatment in liver metastatic colorectal cancer

BACKGROUND

The relationship between the mean absorbed dose delivered to the tumour and the outcome in liver metastases from colorectal cancer patients treated with radioembolization has already been presented in several studies. The optimization of the personalized therapeutic activity to be administered is still an open challenge. In this context, how well the 99mTc-MAA SPECT/CT predicts the absorbed dose delivered by radioembolization is essential. This work aimed to analyse the differences between predictive 99mTc-MAA-SPECT/CT and post-treatment 90Y-microsphere PET/CT dosimetry at different levels. Dose heterogeneity was compared voxel-to-voxel using the quality-volume histograms, subsequently used to demonstrate how it could be used to identify potential clinical parameters that are responsible for quantitative discrepancies between predictive and post-treatment dosimetry.

RESULTS

We analysed 130 lesions delineated in twenty-six patients. Dose-volume histograms were computed from predictive and post-treatment dosimetry for all volumes: individual lesion, whole tumoural liver (TL) and non-tumoural liver (NTL). For all dose-volume histograms, the following indices were extracted: D90, D70, D50, Dmean and D20. The results showed mostly no statistical differences between predictive and post-treatment dosimetries across all volumes and for all indices. Notably, the analysis showed no difference in terms of Dmean, confirming the results from previous studies. Quality factors representing the spread of the quality-volume histogram (QVH) curve around 0 (ideal QF = 0) were determined for lesions, TL and NTL. QVHs were classified into good (QF < 0.18), acceptable (0.18 ≤ QF < 0.3) and poor (QF ≥ 0.3) correspondence. For lesions and TL, dose- and quality-volume histograms are mostly concordant: 69% of lesions had a QF within good/acceptable categories (40% good) and 65% of TL had a QF within good/acceptable categories (23% good). For NTL, the results showed mixed results with 48% QF within the poor concordance category. Finally, it was demonstrated how QVH analysis could be used to define the parameters that predict the significant differences between predictive and post-treatment dose distributions.

CONCLUSION

It was shown that the use of the QVH is feasible in assessing the predictive value of 99mTc-MAA SPECT/CT dosimetry and in estimating the absorbed dose delivered to liver metastases from colorectal cancer via 90Y-microspheres. QVH analyses could be used in combination with DVH to enhance the predictive value of 99mTc-MAA SPECT/CT dosimetry and to assist personalized activity prescription.

Radiation dose-painting with protons vs. photons for head-and-neck cancer

Background: Dose-painting has recently been investigated in early-phase trials in head-and-neck cancer (HNC) with the aim of improving local tumor control. At the same time proton therapy has been reported as potentially capable of decreasing toxicity. Here, we investigate whether protons could be applied in a dose-painting setting by comparing proton dose distributions with delivered photon plans from a phase-I trial of FDG-PET based dose-painting at our institution.Material and methods: Eleven oropharynx (5), hypopharynx (2) and larynx cancer (4) patients from the recently conducted phase I trial were used for comparison of proton and photon dose-painting techniques. Robust optimization (3.5%/3 mm) was used for proton plans. Plan robustness and difference in dose metrics to targets and organs at risk were evaluated.Results: The proton plans met target dose constraints, while having lower non-target dose than photon plans (body-minus-CTV, mean dose 3.9 Gy vs 7.2 Gy, p = .004). Despite the use of robust proton planning for plan max dose, photon plan max doses were more robust (p = .006). Max dose to medulla, brainstem and mandible were lower in the proton plans, while there was no significant difference in mean dose to submandibular- and parotid glands.Conclusion: Proton dose-painting for HNC seems feasible and can reduce the non-target dose overall, however not significantly to certain organs close to the target, such as the salivary glands. Max dose in proton plans had a lower robustness compared to photons, requiring caution to avoid unintended hot spots in consideration of the risk of mucosal toxicity.

Impact of tumour motion compensation and delineation methods on FDG PET-based dose painting plan quality for NSCLC radiation therapy

INTRODUCTION

To quantitatively estimate the impact of different methods for both boost volume delineation and respiratory motion compensation of [18F] FDG PET/CT images on the fidelity of planned non-uniform 'dose painting' plans to the prescribed boost dose distribution.

METHODS

Six locally advanced non-small cell lung cancer (NSCLC) patients were retrospectively reviewed. To assess the impact of respiratory motion, time-averaged (3D AVG), respiratory phase-gated (4D GATED) and motion-encompassing (4D MIP) PET images were used. The boost volumes were defined using manual contour (MANUAL), fixed threshold (FIXED) and gradient search algorithm (GRADIENT). The dose painting prescription of 60 Gy base dose to the planning target volume and an integral dose of 14 Gy (total 74 Gy) was discretized into seven treatment planning substructures and linearly redistributed according to the relative SUV at every voxel in the boost volume. Fifty-four dose painting plan combinations were generated and conformity was evaluated using quality index VQ0.95-1.05, which represents the sum of planned dose voxels within 5% deviation from the prescribed dose. Trends in plan quality and magnitude of achievable dose escalation were recorded.

RESULTS

Different segmentation techniques produced statistically significant variations in maximum planned dose (P < 0.02), as well as plan quality between segmentation methods for 4D GATED and 4D MIP PET images (P < 0.05). No statistically significant differences in plan quality and maximum dose were observed between motion-compensated PET-based plans (P > 0.75). Low variability in plan quality was observed for FIXED threshold plans, while MANUAL and GRADIENT plans achieved higher dose with lower plan quality indices.

CONCLUSIONS

The dose painting plans were more sensitive to segmentation of boost volumes than PET motion compensation in this study sample. Careful consideration of boost target delineation and motion compensation strategies should guide the design of NSCLC dose painting trials.

Functional lung avoidance and response-adaptive escalation (FLARE) RT: Multimodality plan dosimetry of a precision radiation oncology strategy

PURPOSE

Nonsmall cell lung cancer (NSCLC) patient radiation therapy (RT) is planned without consideration of spatial heterogeneity in lung function or tumor response. We assessed the dosimetric and clinical feasibility of functional lung avoidance and response-adaptive escalation (FLARE) RT to reduce dose to [^99m Tc]MAA-SPECT/CT perfused lung while redistributing an escalated boost dose within [¹⁸ F]FDG-PET/CT-defined biological target volumes (BTV).

METHODS

Eight stage IIB-IIIB NSCLC patients underwent FDG-PET/CT and MAA-SPECT/CT treatment planning scans. Perfused lung objectives were derived from scatter/collimator/attenuation-corrected MAA-SPECT uptake relative to ITV-subtracted lung to maintain < 20 Gy mean lung dose (MLD). Prescriptions included 60 Gy to the planning target volume (PTV) and concomitant boost of 74 Gy mean to biological target volumes (BTV = GTV + PET gradient segmentation) scaled to each BTV voxel by relative FDG-PET SUV. Dose-painting-by-numbers prescriptions were integrated into commercial treatment planning systems via uptake threshold discretization. Dose constraints for lung, heart, cord, and esophagus were defined. FLARE RT plans were optimized with volumetric modulated arc therapy (VMAT), proton pencil beam scanning (PBS) with 3%-3 mm robust optimization, and combination of PBS (avoidance) plus VMAT (escalation). The high boost dose region was evaluated within a standardized SUV_peak structure. FLARE RT plans were compared to reference VMAT plans. Linear regression between radiation dose to BTV and normalized FDG PET SUV at every voxel was conducted, from which Pearson linear correlation coefficients and regression slopes were extracted. Spearman rank correlation coefficients were estimated between radiation dose to lung and normalized SPECT uptake. Dosimetric differences between treatment modalities were evaluated by Friedman nonparametric paired test with multiple sampling correction.

RESULTS

No unacceptable violations of PTV and normal tissue objectives were observed in 24 FLARE RT plans. Compared to reference VMAT plans, FLARE VMAT plans achieved a higher mean dose to BTV (73.7 Gy 98195. 61.3 Gy), higher mean dose to SUV_peak (89.7 Gy vs. 60.8 Gy), and lower mean dose to highly perfused lung (7.3 Gy vs. 14.9 Gy). These dosimetric gains came at the expense of higher mean heart dose (9.4 Gy vs. 5.8 Gy) and higher maximum cord dose (50.1 Gy vs. 44.6 Gy) relative to the reference VMAT plans. Between FLARE plans, FLARE VMAT achieved higher dose to the SUV_peak ROI than FLARE PBS (89.7 Gy vs. 79.2 Gy, P = 0.01), while FLARE PBS delivered lower dose to lung than FLARE VMAT (11.9 Gy vs. 15.6 Gy, P < 0.001). Voxelwise linear dose redistribution slope between BTV dose and FDG PET uptake was higher in magnitude for FLARE PBS + VMAT (0.36 Gy per %SUV_max ) compared to FLARE VMAT (0.27 Gy per %SUV_max ) or FLARE PBS alone (0.17 Gy per %SUV_max ).

CONCLUSIONS

FLARE RT is clinically feasible with VMAT and PBS. A combination of PBS for functional lung avoidance and VMAT for FDG PET dose escalation balanced target and normal tissue objective tradeoffs. These results provide a technical platform for testing of FLARE RT safety and efficacy within a precision radiation oncology trial.

Hypoxia imaging and radiotherapy: bridging the resolution gap

Oxygen distribution is a major determinant of treatment success in radiotherapy, with well-oxygenated tumour regions responding by up to a factor of three relative to anoxic volumes. Conversely, tumour hypoxia is associated with treatment resistance and negative prognosis. Tumour oxygenation is highly heterogeneous and difficult to measure directly. The recent advent of functional hypoxia imaging modalities such as fluorine-18 fluoromisonidazole positron emission tomography have shown promise in non-invasively determining regions of low oxygen tension. This raises the prospect of selectively increasing dose to hypoxic subvolumes, a concept known as dose painting. Yet while this is a promising approach, oxygen-mediated radioresistance is inherently a multiscale problem, and there are still a number of substantial challenges that must be overcome if hypoxia dose painting is to be successfully implemented. Current imaging modalities are limited by the physics of such systems to have resolutions in the millimetre regime, whereas oxygen distribution varies over a micron scale, and treatment delivery is typically modulated on a centimetre scale. In this review, we examine the mechanistic basis and implications of the radiobiological oxygen effect, the factors influencing microscopic heterogeneity in tumour oxygenation and the consequent challenges in the interpretation of clinical hypoxia imaging (in particular fluorine-18 fluoromisonidazole positron emission tomography). We also discuss dose-painting approaches and outline challenges that must be addressed to improve this treatment paradigm.

Robustness of quantitative hypoxia PET image analysis for predicting local tumor control

BACKGROUND

Previous studies suggested the maximum tumor to background ratio (TBRmax) in FMISO PET images as a potentially predictive parameter for local control after radio-chemotherapy (CRT) in head and neck squamous cell carcinomas (HNSCC). However, different TBRmax thresholds for stratification were reported, implying that a common threshold cannot readily be used among different institutions without the risk of reducing prediction accuracy. Therefore, this study investigated the robustness of using a common pre-defined TBRmax, simulating a multicenter clinical trial.

MATERIAL AND METHODS

FMISO PET/CT was performed four hours post-injection in 22 patients with advanced HNSCC in a phase II FMISO dose escalation study. PET background regions of interest (ROIs) were manually defined in deep neck muscles. TBRmax was calculated as the mean of the highest-valued voxels within the high risk RT planning target volume. Its predictive power with respect to local control was tested, classifying patients using median TBRmax as threshold. The influence of systematically varying quantification between institutions was studied in silico by applying offsets of ± 10% and ± 20% to the TBRmax of all patients, while the threshold remained constant. The effect was analyzed using a receiver operating characteristic (ROC). True positive and false positive rates (TPR/FPR) as well as positive and negative predictive values (PPV/NPV) were evaluated.

RESULTS

For the reference condition without an offset the median TBRmax was 2.0 (1.4-3.5). Patients were classified using this threshold and TPR = 0.7, FPR = 0.4, PPV = 0.5 and NPV = 0.8 were observed. Accuracy declined with increasing offsets. Negative offsets of -10% and -20% resulted in TPR = 0.43 and 0.14, FPR = 0.20 and 0.13, PPV = 0.50 and 0.33 and NPV = 0.75 and 0.68, respectively. Positive offsets of + 10% and + 20% resulted in TPR = 1.00 and 1.00, FPR = 0.53 and 0.67, PPV = 0.47 and 0.41 and NPV = 1.00 and 1.00, respectively.

CONCLUSIONS

Using a common pre-defined TBRmax threshold in multicenter trials requires careful standardization and harmonization of all steps from patient preparation to image analysis. Our results indicate that TBRmax should deviate less than 10% from reference conditions (absolute value in this dataset ± 0.2). This conclusion likely applies to all low contrast nitroimidazole hypoxia PET tracers.