The use of PET images for radiotherapy treatment planning: an error analysis using radiobiological endpoints

There is significant current interest in the use of biological image guidance in radiotherapy planning. In lung-cancer treatment, tumor motion due to respiration is known to be a limitation. This is particularly true in PET, where image data are collected over a number of minutes. An in-house-developed 4D PET acquisition mode is described and an analysis of the effects of acquisition parameters on the reconstructed image quality is presented. The potential impact of the resulting biological image quality on radiotherapy planning is then quantified in terms of tumor control probability (TCP).

Data were acquired using a human torso phantom comprised of a hot '8F-filled spheroidal "tumor" (40 mm in diameter) suspended in an air-filled "lung" cylinder and surrounded by a warm 18F-filled background. Two different sphere-to-background (S/B) ratios were used. The tumor was connected to a 3-axis computer-controlled motion stage and could be moved during PET data acquisition. Images were acquired with a range of count statistics, motion blurring, and CT attenuation correction (CTAC) misalignment. Four simple models were proposed for the assignment of clonogenic cell density according to the voxel value. The impact of image artifacts was then assessed by calculating the TCP, which is the probability that no clonogenic tumor cell remains after a given dose of radiation. TCP was calculated for a uniform dose distribution in the tumor.

Reduced count statistics and misaligned CTAC images had the most detrimental impact on the image fidelity. It was found that in both cases the images became less intense, demonstrated by smaller number of voxels at the maximum values. The maximum TCP difference between images with the least and most noise was 3.4% (S/B=3), and with weakest and strongest CT misalignment artifacts, it was 3.2% (S/B = 10). Motion blurring only contributed weakly to the TCP imprecision at 1.7% (S/B=10) between best- and worst-case images. However, the model-calculated TCP showed increasing differences from the ground truth as the complexity of the model increased [maximum difference of approximately 8% (model 3)], which could be attributed to the partial volume effect.

Based on the results of this study, it is believed that simple techniques of biologically guided radiotherapy planning for lung cancer should be feasible at intermediate contrast levels (tumor-to-background ratio of approximately 10) with the clinically achievable image quality.