2303 - VQ-VAE-Based Generative Model for Predicting Anatomical Changes in Radiotherapy

02:30pm - 04:00pm PT

Hall F

Screen: 29

POSTER

Presenter(s)

Zhen Zhang, MD - Department of Radiation Oncology, Fudan University Shanghai Cancer Center, Shanghai,

Z. Zhang¹, Y. Zou¹, J. Wang², and W. Hu²; ¹Department of Radiation Oncology, Fudan University Shanghai Cancer Center, ???, China, ²Department of Radiation Oncology, Fudan University Shanghai Cancer Center, Shanghai, Shanghai, China

Purpose/Objective(s): Anatomical variations during radiotherapy fractions can lead to deviations in radiation delivery. This study proposes a VQ-VAE-based generative model to predict variations in nasopharyngeal cancer patients.

Materials/Methods: The model uses the Vector Quantized Variational Autoencoder (VQ-VAE) framework combined with Adaptive Instance Normalization (AdaIN). The VQ-VAE encodes anatomical structures from planning CT images, while a convolutional neural network (CNN) extracts anatomical variations. AdaIN then guides the VQ-VAE to generate daily CT images reflecting potential anatomical changes. The model was trained and validated on 520 CT scans from 90 nasopharyngeal cancer patients and tested using 102 CT scans from 18 patients. The quality of the generated images was evaluated through visual inspection, while the model’s accuracy was assessed by comparing the predicted and actual volumes of the parotid and submandibular glands at both individual and population levels.

Results: For individual patients, the average p-value of the KS and KM tests for OAR volume distributions is 0.352. The results indicate that the majority of the generated organ-at-risk (OAR) distributions show no statistically significant differences compared to the distributions observed in daily clinical CT images (P > 0.05). At the population level, the mean and standard deviation of ROI volumes (parotid glands: 26.9 ± 2.1 cm³; submandibular glands: 7.0 ± 0.71 cm³) showed smaller discrepancies from the ground truth values (parotid glands: 29.5 ± 3.2 cm³; submandibular glands: 7.2 ± 0.67 cm³) compared to the differences observed with the DAM model.

Conclusion: The VQ-VAE model can predict potential anatomical changes during radiotherapy based on the patient’s planning CT and have the potential to be used for developing personalized treatment plans and making timely treatment adjustments.