Room: AAPM ePoster Library
Automated segmentation of esophagus is critical in image guided/adaptive radiotherapy of lung cancer to minimize radiation-induced toxicities such as acute esophagitis. We developed a semantic physics-based data augmentation method for segmenting esophagus in both planning CT (pCTs) and cone-beam CTs (CBCTs) using 3D convolutional neural networks.
The study included 60 lung patients treated via IMRT and had weekly CBCTs. First, week1 CBCTs were deformably registered to their pCTs using regularized diffeomorphic registration. Then, 7 variations of scatter artifacts/noise were extracted from the week1 CBCTs using power-law adaptive histogram equalization that contained the highest to the smoothest frequency components. Extracted CBCT artifacts were added to their corresponding pCT, and 2D projections were generated from the artifact-induced pCTs. The projections were reconstructed using iterative ordered-subset simultaneous algebraic reconstruction technique to generate pseudo CBCTs and were evaluated against their ground-truth CBCT using structure similarity index (SSIM) & root mean square error (RMSE). The pseudo CBCTs were split into 70/30 training (n=294) and testing cases (n=90) and fed to a 3D-UNet for esophagus segmentation using pCT esophagus contours as ground-truth. The model was externally validated on the weekly CBCTs and pCTs using dice coefficient and volume differences between the physician-contoured and UNet-segmented esophagus.
The best reconstructed pseudo CBCTs had average SSIM=0.89 and RMSE=0.05. Average dice for segmenting pseudo CBCTs, weekly CBCTs, and pCT were 0.74±0.03, 0.71±0.05, and 0.77±0.04, respectively, considerably better than published AAPM segmentation challenge results of 0.72. Average volume differences were 2.6±1.4cc (pseudo-CBCT), 3.4±1.6cc (weekly CBCTs) and 2.3±1.5cc (pCTs).
3D-UNet model trained on more realistic artifact-induced pCTs, could segment esophagus on both weekly CBCTs and pCTs with high accuracy for longitudinal imaging studies. The model has a potential to segment any OAR on CBCT/pCT and therefore, can be used as a cross-modality segmentation tool to provide image guidance.
Funding Support, Disclosures, and Conflict of Interest: Master research agreement between Memorial Sloan Kettering Cancer Center and Varian