ESTRO 2024 - Abstract Book

S4201

Physics - Intra-fraction motion management and real-time adaptive radiotherapy

ESTRO 2024

Daniel A Low 1 , Michael V Lauria 1 , Minji V Kim 1 , Dylan O'Connell 1 , Yi Lao 1 , Drew Moghanaki 1 , Alan Lee 1 , Ricky Savjani 1 , Jonathan Goldin 2 , Igor Barjaktarevic 3 , Claudia Miller 1 , Louise Naumann 1 1 UCLA, Radiation Oncology, Los Angeles, USA. 2 UCLA, Radiology, Los Angeles, USA. 3 UCLA, Pulmonology, Los Angeles, USA

Purpose/Objective:

To describe the clinical model-based CT (5DCT) experience at UCLA for the first 156 lung cancer patients.

Material/Methods:

Model-based CT has been proposed as an alternative to 4DCT and our group developed a clinical workflow that entered service in 2019 and has since been used on 156 lung cancer patients (169 scans). The approach we have developed assumes that breathing motion can be described as a function of two breathing surrogates 1 . The first is the breathing amplitude which is responsible for describing the general inflation and deflation of the lungs. The second is the breathing rate, which is a surrogate for pressure imbalances during breathing and is responsible for describing motion hysteresis. Because we use two surrogates, we term this approach 5DCT 1 . The 5DCT workflow consists of 1) Conducting 25 low-dose free-breathing helical CT scans with simultaneous breathing surrogate measurement 2,3 , 2) Deformably registering each of the 24 scans to the first scan, which is used as the reference scan, 3) Associating each free-breathing scan slice with its associated breathing amplitude and rate, 4) Fitting a universal equation of breathing motion on a voxel by voxel basis 1 , and calculate the model residual as an indication of workflow precision, 5) Providing our clinic with 8 reconstructed CT scans at 8 breathing amplitudes and their associated rates to mimic the clinical 4DCT output, and 6) Use the model to reconstruct the original 25 CT scans and compare them to the actual scans to estimate end-to-end workflow accuracy 4 . For each patient, we evaluated the quality of the 5DCT workflow steps including original CT images for artifacts, image registration, and workflow accuracy. We also evaluated breathing regularity. These were described on a 1 to 4 scale, where 1 was excellent or regular, and 4 was unusable or extremely irregular for the three quality and breathing regularity assessments, respectively. Spearman correlation coefficients were calculated to examine the associations between the grades assigned to the validation scan reconstructions and variables such as breathing rate, irregularities, image artifacts, and registration errors. We also tabulated the fraction of patients for which deformation failed, causing the 5DCT scans to be clinically unusable.

Results:

For context, figure 1 compares 4DCT and 5DCT scans for the same patient during mid-inspiration. Among the 156 clinical 5DCT patients, workflow accuracy was correlated best with model residual and image registration with Pearson and Spearman correlation coefficients of 0.656 (p<0.001) and 0.500 (p<0.001), respectively. 32 patients had breathing irregularities graded as a 4, but only 9 had workflow accuracy graded as a 4, and the Spearman correlation coefficient between these two was only 0.260 (p<0.001). Finally the correlation between image artifacts and overall workflow quality was also relatively low at 0.301 (p<0.001). These correlations indicate that the proposed approach is least sensitive to breathing irregularities and most sensitive to image registration errors. Due, we think, primarily to image registration failures, the 5DCT results were unusable for 21 scans (12%) and the fast-helical scans were used to estimate tumor motion using either a "mega MIP" or scrolling through the scans to evaluate motion.

Figure 1: Comparison between clinical 4DCT and 5DCT scans for the same patient.