Room: Exhibit Hall | Forum 5
Purpose: By obtaining the energy dependence of photoelectric and Compton interactions, dual-energy CT (DECT) can be used to derive a number of parameters based on physics properties, such as relative stopping power map (RSPM). The accuracy of DECT-derived parametric maps relies on image noise levels and the severity of artifacts. Sub-optimal image quality may degrade the accuracy of physics-based mapping techniques and affect subsequent processing for clinical applications. In this study, we propose a deep-learning-based method to accurately generate RSPM based on the virtual monoenergetic images as an alternative to physics-based dual-energy approaches.
Methods: For the training target of our deep-learning model, we manually segmented head-and-neck DECT images into brain, bone, fat, soft-tissue, lung and air, and then assigned different RSP values into the corresponding tissue types to generate a reference RSPM. We proposed to integrate a residual block concept into a cycle-consistent generative adversarial network framework to learn the nonlinear mapping between 70keV/140keV monoenergetic images and reference RSPM. We evaluated the proposed method with 18 head-and-neck cancer patients. Mean absolute error (MAE) and mean error (ME) were used to quantify the differences between the generated and reference RSPM.
Results: The average MAE between generated and reference RSPM was 3.1Â±0.4 % and the average ME was 1.5Â±0.5 % for all patients. Compared to the physics-based method, the proposed method could significantly improve RSPM accuracy and had comparable computational efficiency after training.
Conclusion: We developed a novel learning-based method to effectively capture the relationship between DECT data of tissue substitutes and reference RSPM. We subsequently used the method to generate accurate RSPM and demonstrated its reliability on head-and-neck patients. The proposed deep-learning-based approach has the potential advantages of producing unbiased and robust RSPM for proton dose calculation.
Funding Support, Disclosures, and Conflict of Interest: NIH R01 215718
Not Applicable / None Entered.