Radiation and Cancer Physics
SS 33 - Physics 10- Machine Learning for Planning and Segmentation
195 - An Image-Based Framework for Individualizing Radiotherapy Dose
Wednesday, September 18
12:15 PM - 12:25 PM
Location: Room W185
Mohamed Abazeed, MD, PhD
Cleveland Clinic Taussig Cancer Institute
Cleveland Clinic: Assistant Professor: Employee
Bayer: Honoraria: Research Grants: Travel Expenses;
Siemens Healthcare: Research Grants
An Image-Based Framework for Individualizing Radiotherapy Dose
B. Lou1, S. Doken2, T. Zhuang3, D. Wingerter2, N. Mistry4, L. Ladic5, A. Kamen5, and M. Abazeed3; 1Siemens Healthineers, Malvern, PA, United States, 2Cleveland Clinic, Cleveland, OH, 3Department of Radiation Oncology, Taussig Cancer Institute, Cleveland Clinic, Cleveland, OH, 4Siemens Healthineers, Knoxville, TN, 5Siemens Healthineers, Princeton, NJ
Purpose/Objective(s): Radiotherapy continues to be delivered uniformly without taking into account individual tumor characteristics. We queried the lung computed tomography (CT)-derived feature space to identify radiation sensitivity parameters that can predict treatment failure and hence guide the individualization of radiotherapy dose.
Materials/Methods: We used a cohort-based registry of 849 patients with cancer in the lung treated with high-dose radiotherapy. We input pre-therapy lung CT images into a deep neural network, Deep Profiler, to generate an image fingerprint that primarily predicts treatment outcomes and secondarily approximates classical radiomic features. Deep Profiler was combined with electronic health records (EHR) data to derive an individualized radiation dose, iGray, a patient-specific dose that reduces treatment failure probability to <5%.
Results: Patients with high Deep Profiler scores fail radiation at a significantly higher rate than in those with low scores. The 3-year cumulative incidences of local failure were 20.3% (95% CI: 16.0-24.9) and 5.7% (95% CI: 3.5-8.8), respectively. Deep Profiler independently predicted local failure (hazard ratio 1.65, 95% 1.02-2.66, p=.042). Models that included Deep Profiler and EHR predicted treatment failures with a concordance index of 0.721, a significant improvement compared to classical radiomics or clinical variables alone (p=4.64×10-14 and 4.57×10-22, respectively). iGray had a wide dose range (21.1-277 Gy, BED), suggested dose reduction in 23.3% of patients and can be safely delivered in the majority of cases. Voxel saliency maps indicate that 37.4% of the voxels that are most deterministic for treatment failure localize outside of the physician-contoured tumor volume.
Conclusion: Our image-based deep learning framework is the first opportunity to use medical images to individualize radiation dose delivery.
Author Disclosure: B. Lou: None. S. Doken: None. T. Zhuang: None. D. Wingerter: None. M. Abazeed: Research Grant; Bayer, Siemens Healthcare. Honoraria; Bayer. Travel Expenses; Bayer.