Purpose: Use of CT scanners for oncology treatment planning has received significant interest. Since the patient needs to be positioned in a similar fashion as in a therapy machine, a part of the patient is often placed outside the field-of-view (FOV) defined by the CT scanner. The patient anatomy outside of the FOV may lie in the treatment beam path and therefore must be considered when calculating X-ray attenuation and dose delivery. In addition, truncated projections often produce image artifacts and make attenuation estimation difficult. We propose a reconstruction algorithm that allows adequate estimation of the object outside the FOV.
Methods and Materials: Since the total attenuation of each ideal projection in a parallel sampling geometry remains constant over views, we can accurately estimate the sum of missing projection data. To overcome the small fluctuation due to non-perfect calibrations and patient motion, we use the projections of two neighboring non-truncated views as the basis. The missing attenuation for each view is the difference between the linearly interpolated total attenuation and the actual measurement. Next, we use the magnitude and slope of the projection samples at the truncation to estimate the size and the location of a water cylinder that can best fit to the projection data. Additional expansion and contraction of the fitted projection takes place based on the difference between the total attenuation of the fitted missing projection and the previously calculated projection. Continuity constraints are placed on the fitting parameters.
Results: Extensive phantom and clinical experiments were conducted to validate our approach. For quantitative analysis, we positioned the phantoms from 0cm to 8cm outside the 50cm FOV at 1cm increments. Eight different phantoms were used: 48cm poly, 35cm poly, 20cm GE QA resolution phantom, 20cm QA low-contrast phantom, 20 cm QA water phantom, 20 cm linearity phantom, 5 in. water phantom, and a thorax phantom. Results show that CT number accuracy is fully recovered inside the 50cm FOV in all cases. The maximum amount of CT number deviation outside FOV is about 40HU. The shape of the truncated object has been recovered to an accuracy of a few millimeters.
Conclusion: We present an algorithmic approach to extend the FOV of a CT scanner beyond the hardware limit. Extensive phantom and clinical experiments have shown that the algorithm can successfully restore images inside the FOV without residual truncation artifacts. Regions outside the FOV is adequate for oncology applications. (J.H., E.C., B.G., A.H., S.M., T.J.M. are employees of GE Medical Systems.)