Purpose: To investigate the feasibility of automatic segmentation of head and neck (H&N) tumors in kV images using simulated projections. Success will lead to the ability to track tumor motion during H&N radiotherapy with the goal of removing the need for the thermoplastic mask for radiotherapy.
Methods: CT scans of two patients were used from the HNSCC database from the Cancer Imaging Archive. For each patient, 3600 evenly spaced digitally reconstruction radiographs (DRRs) were generated (0.1° projection angle difference) from the CT scan. For each DRR the ground truth tumor location was forward projected from the planning CT and corresponding structures file. A Cycle Generation Adversarial Network (cGAN) was trained to detect the tumour location in each DRR. The testing data consisted of 360 DRRs from a rigidly transformed copy of the CT volume, with the projection angles of the DRRs being taken from a typical CBCT scan for H&N patients. The CT scan was rotated by 3° and then translated by 3mm in the AP, LR and SI directions. The estimated tumor location from the cGAN was compared with the ground truth for each patient by measuring the centroid error and the DICE similarity coefficient (DSC).
Results: The mean(±Standard deviation) DSC coefficients for patients 1 and 2 were 0.89±0.02, 0.90±0.02, and for all patients the DSC was 0.89±0.02. The mean(±Standard deviation) absolute centroid error for patients 1 and 2 were 1.43±0.76mm and 1.54±0.88mm, and for all patients was 1.49±0.82mm.
Conclusion: The results in this study demonstrate potential for tracking the tumor location for head and neck cancers in silico. Further development of this technology will pave the way for the removal of the need for thermoplastic masks for treating head and neck cancers during radiotherapy.
Funding Support, Disclosures, and Conflict of Interest: This project received funding support from Cancer Australia
Target Localization, Image Analysis
Not Applicable / None Entered.