Dose-Incorporated Deep Ensemble Learning for Improving Brain Metastasis Stereotactic Radiosurgery Outcome Prediction.

PURPOSE: To develop a novel deep ensemble learning model for accurate prediction of brain metastasis (BM) local control outcomes after stereotactic radiosurgery (SRS). METHODS AND MATERIALS: A total of 114 brain metastases (BMs) from 82 patients were evaluated, including 26 BMs that developed biopsy-confirmed local failure post-SRS. The SRS spatial dose distribution (Dmap) of each BM was registered to the planning contrast-enhanced T1 (T1-CE) magnetic resonance imaging (MRI). Axial slices of the Dmap, T1-CE, and planning target volume (PTV) segmentation (PTVseg) intersecting the BM center were extracted within a fixed field of view determined by the 60% isodose volume in Dmap. A spherical projection was implemented to transform planar image content onto a spherical surface using multiple projection centers, and the resultant T1-CE/Dmap/PTVseg projections were stacked as a 3-channel variable. Four Visual Geometry Group (VGG-19) deep encoders were used in an ensemble design, with each submodel using a different spherical projection formula as input for BM outcome prediction. In each submodel, clinical features after positional encoding were fused with VGG-19 deep features to generate logit results. The ensemble's outcome was synthesized from the 4 submodel results via logistic regression. In total, 10 model versions with random validation sample assignments were trained to study model robustness. Performance was compared with (1) a single VGG-19 encoder, (2) an ensemble with a T1-CE MRI as the sole image input after projections, and (3) an ensemble with the same image input design without clinical feature inclusion. RESULTS: The ensemble model achieved an excellent area under the receiver operating characteristic curve (AUCROC: 0.89 ± 0.02) with high sensitivity (0.82 ± 0.05), specificity (0.84 ± 0.11), and accuracy (0.84 ± 0.08) results. This outperformed the MRI-only VGG-19 encoder (sensitivity: 0.35 ± 0.01, AUCROC: 0.64 ± 0.08), the MRI-only deep ensemble (sensitivity: 0.60 ± 0.09, AUCROC: 0.68 ± 0.06), and the 3-channel ensemble without clinical feature fusion (sensitivity: 0.78 ± 0.08, AUCROC: 0.84 ± 0.03). CONCLUSIONS: Facilitated by the spherical image projection method, a deep ensemble model incorporating Dmap and clinical variables demonstrated excellent performance in predicting BM post-SRS local failure. Our novel approach could improve other radiation therapy outcome models and warrants further evaluation.

Independent Monte Carlo dose calculation identifies single isocenter multi-target radiosurgery targets most likely to fail pre-treatment measurement.

PURPOSE: For individual targets of single isocenter multi-target (SIMT) Stereotactic radiosurgery (SRS), we assess dose difference between the treatment planning system (TPS) and independent Monte Carlo (MC), and demonstrate persistence into the pre-treatment Quality Assurance (QA) measurement. METHODS: Treatment plans from 31 SIMT SRS patients were recalculated in a series of scenarios designed to investigate sources of discrepancy between TPS and independent MC. Targets with > 5% discrepancy in DMean[Gy] after progressing through all scenarios were measured with SRS MapCHECK. A matched pair analysis was performed comparing SRS MapCHECK results for these targets with matched targets having similar characteristics (volume & distance from isocenter) but no such MC dose discrepancy. RESULTS: Of 217 targets analyzed, individual target mean dose (DMean[Gy]) fell outside a 5% threshold for 28 and 24 targets before and after removing tissue heterogeneity effects, respectively, while only 5 exceeded the threshold after removing effect of patient geometry (via calculation on StereoPHAN geometry). Significant factors affecting agreement between the TPS and MC included target distance from isocenter (0.83% decrease in DMean[Gy] per 2 cm), volume (0.15% increase per cc), and degree of plan modulation (0.37% increase per 0.01 increase in modulation complexity score). SRS MapCHECK measurement had better agreement with MC than with TPS (2%/1 mm / 10% threshold gamma pass rate (GPR) = 99.4 ± 1.9% vs. 93.1 ± 13.9%, respectively). In the matched pair analysis, targets exceeding 5% for MC versus TPS also had larger discrepancies between TPS and measurement with no GPR (2%/1 mm / 10% threshold) exceeding 90% (71.5% ± 16.1%); whereas GPR was high for matched targets with no such MC versus TPS difference (96.5% ± 3.3%, p = 0.01). CONCLUSIONS: Independent MC complements pre-treatment QA measurement for SIMT SRS by identifying problematic individual targets prior to pre-treatment measurement, thus enabling plan modifications earlier in the planning process and guiding selection of targets for pre-treatment QA measurement.

Image-guided and intensity-modulated radiosurgery for patients with spinal metastasis.

BACKGROUND: Radiosurgery can deliver a single, large radiation dose to a localized tumor using a stereotactic approach and hence, requires accurate and precise delivery of radiation to the target. Of the extracranial organ targets, the spine is considered a suitable site for radiosurgery, because there is minimal or no breathing-related organ movement. The authors studied spinal radiosurgery in patients with spinal metastases to determine its accuracy and precision. METHODS: The spinal radiosurgery program was based on an image-guided and intensity-modulated, shaped-beam radiosurgical unit. It is equipped with micromultileaf collimators for beam shaping and radiation intensity modulation and with a noninvasive, frameless positioning device that uses infrared, passive marker technology together with corroborative image fusion of the digitally reconstructed image from computed tomography (CT) simulation and orthogonal X-ray imagery at the treatment position. These images were compared with the port films that were taken at the time of treatment to determine the accuracy of the isocenter position. Clinical feasibility was tested in 10 patients who had spinal metastasis with or without spinal cord compression. The patients were treated with fractionated external beam radiotherapy followed by single-dose radiosurgery as a boost (6-8 grays) to the most involved portion of the spine or to the site of spinal cord compression. RESULTS: The accuracy for the isocenter was within 1.36 mm +/- 0.11 mm, as measured by image fusion of the digitally reconstructed image from CT simulation and the port film. Clinically, the majority of patients had prompt pain relief within 2-4 weeks of treatment. Complete and partial recovery of motor function also was achieved in patients with spinal cord compression. The radiation dose to the spinal cord was minimal. The maximum dose of radiation to the anterior edge of the spinal cord within a transverse section, on average, was 50% of the prescribed dose. There was no acute radiation toxicity detected clinically during the mean follow-up of 6 months. CONCLUSIONS: Image-guided, shaped-beam spinal radiosurgery is accurate and precise. Rapid clinical improvement of pain and neurologic function also may be achieved. The results indicate the potential of spinal radiosurgery in the treatment of patients with spinal metastasis, especially those with solitary sites of spine involvement, to increase the prospects of long-term palliation.