Session Item

To present the first clinical implementation of a model-based CT workflow for lung cancer radiation therapy.

This work was motivated by the lack of quantitation and the sorting-induced artifacts of commercial 4DCT.  The model-based CT workflow starts with the acquisition of 25 low-mAs fast helical CT scans (the first termed the reference image) using a 64-slice CT scanner with simultaneous abdomen-based breathing surrogate measurement.  The breathing surrogate amplitude is assigned to each CT slice.  Deformable image registration is used to determine the voxel-voxel motion and a breathing motion model fit to each voxel. For our system, the model uses breathing amplitude and rate as the time-dependent variables, hence is termed 5DCT.  The model is subsequently used to deform the reference image to a user-selected breathing amplitude.  The model residuals are used to describe overall process quality, while the original CT scans are also reconstructed to describe overall process accuracy. For the clinic, we replace the 8 phase-based CT scans with model-built scans at 8 amplitudes corresponding to 8 breathing amplitude percentiles. 
We evaluated the first 13 clinical patients to determine the impact and quality of the new workflow, including analyzing breathing irregularity and the model residuals.  

Seven of the 13 patients had irregular breathing, determined by examining the breathing trace during the CT scan acquisition.  These seven had a combination of varying breathing amplitude and baseline drift. Two patients had wildly varying amplitudes, three had significant drift, and the last had more breath-to-breath erratic breathing.  The corresponding motion model residuals were compared between regular and irregular breathing patients, with mean residual values of 1.23 mm and 1.45 mm, respectively, but the differences were not statistically significant (p = 0.27).  The 95^th percentile of the residuals was used to characterize outliers, which were 2.41 mm, and 2.35 mm for the normal and abnormal breathing patterns, respectively.
Figure 1 shows an example of regular (patient 2) and irregular (patient 5) breathing patterns.  Table 1 shows the statistical analysis of the mean and 95^th percentile model residual errors.  While the mean values of the residuals did not correlate to whether the breathing amplitude was irregular, the two patients with the largest residuals both exhibited irregular breathing patterns.  Note that the irregular breathing patient shown in Figure 1 had small mean and 95^th percentile residuals.

The workflow requires that the therapist set up the patient similarly to conventional 4DCT. The overall scan time is 2-3 minutes, and the dose is nearly the same as conventional 4DCT.  The resulting images are sorting artifact free and the model 95^th percentile residual is typically less than 4 mm (regardless of whether the breathing is regular), although we have seen two cases of >4 mm, both of which had highly irregular breathing.