S2317
Physics - Machine learning and AI algorithms
ESTRO 2026
whereas VMAT is preferred for TI >0.014.
image segmentation. arXiv. 2018;1809.10486. Keywords: Generative model, Digital phantom, Virtual trials
Digital Poster 4765
Optimizing Breast Cancer Radiotherapy: AI and Tangential Index for VMAT vs HYBRID Selection ABEL RODRIGUEZ ARANDA 1 , ADRIN ANDRADES 2 1 Medical-Physics, Hospital Universitario de Lanzarote, Arrecife, Spain. 2 Radiotherapy, Las Palmas University Hospital, Las Palmas, Spain Purpose/Objective: In breast cancer (BC) radiotherapy, selecting the optimal technique between VMAT and HYBRID can be challenging. This study introduces two decision- support tools: a new parameter, the Tangential Index (TI), and an artificial intelligence (AI) model. Material/Methods: A retrospective study was conducted on 205 BC patients. For each patient, VMAT and HYBRID plans were generated using Eclipse TPS, selecting the plan that provided the best PTV coverage while respecting OAR dose constraints.The Tangential Index (TI) was defined as the ratio of the breast curvature angle ( α ) to the ipsilateral lung volume (Eq. 1). Breast–heart distance (B-H distance) was also measured as an additional decision parameter. In addition, the breast–heart distance (B-H distance) was measured, as this parameter strongly influences dose distribution and plan selection.A convolutional neural network (CNN) was developed to predict the optimal technique (VMAT, HYBRID, or VMAT/HYBRID for plans with very similar outcomes) by localizing the PTV within axial CT regions of interest (ROIs, see Figure 1). The model was trained on 90% of the dataset, with the remaining 10% used for validation. Data augmentation and VGG16-based transfer learning were applied to enhance feature extraction, increase dataset variability, and reduce overfitting. Results: As shown in the Graph 1, TI values <0.010 strongly favor HYBRID; values between 0.010–0.014 suggest a balance between HYBRID and VMAT, with a slight bias toward HYBRID. TI values >0.014 favor VMAT, reflecting easier compliance with OAR constraints and a higher tangential index.For left breast (LB) patients, B-H distance <0.63 cm favors VMAT due to closer heart proximity; distances between 0.63–1.25 cm favor HYBRID, while distances >1.25 cm strongly favor HYBRID. Overall, HYBRID is optimal for TI <0.010,
Thus, the recommended selection guidelines are:Right breast (RB): TI <0.013 → VMAT; TI <0.010 → HYBRIDLeft breast (LB): TI >0.015 & B-H <0.63 cm → VMAT; TI 0.010–0.015 & B-H 0.73–1.25 cm → either technique; TI <0.010 & B-H >1.25 cm → HYBRIDIn addition, the AI model (outcome in Figure 1) achieved >92% accuracy in training, with prediction accuracies of 87.5% for VMAT, 77.7% for HYBRID, and 44.4% for VMAT/HYBRID. This indicates robust performance despite the limited dataset of 205 patients.
Conclusion: Integrating the Tangential Index (TI) and B-H distance with AI-based predictions provides a robust clinical tool to guide the optimal choice between VMAT and HYBRID in breast cancer radiotherapy. References: Voyant, C., Pinpin, M., Leschi, D. et al. Hybrid VMAT- 3DCRT as breast cancer treatment improvement tool. Sci Rep 13, 23110 (2023). https://doi.org/10.1038/s41598-023-50538-xBi, S., Zhu, R. & Dai, Z. Dosimetric and radiobiological comparison of simultaneous integrated boost radiotherapy for early stage right side breast cancer between three
Made with FlippingBook - Share PDF online