Pankaj Mishra

Respiratory motion during radiotherapy can cause uncertainties in definition of the target volume and in estimation of the dose delivered to the target and healthy tissue. In this paper, we generate volumetric images of the internal patient anatomy during treatment using only the motion of a surrogate signal. Pre-treatment four-dimensional CT imaging is used to create a patient-specific model correlating internal respiratory motion with the trajectory of an external surrogate placed on the chest. The performance of this model is assessed with digital and physical phantoms reproducing measured irregular patient breathing patterns. Ten patient breathing patterns are incorporated in a digital phantom. For each patient breathing pattern, the model is used to generate images over the course of thirty seconds. The tumor position predicted by the model is compared to ground truth information from the digital phantom. Over the ten patient breathing patterns, the average absolute error in the tumor centroid position predicted by the motion model is 1.4 mm. The corresponding error for one patient breathing pattern implemented in an anthropomorphic physical phantom was 0.6 mm. The global voxel intensity error was used to compare the full image to the ground truth and demonstrates good agreement between predicted and true images. The model also generates accurate predictions for breathing patterns with irregular phases or amplitudes.

Accurate understanding and modeling of respiration-induced uncertainties is essential in image-guided radiotherapy. Explicit modeling of the overall lung motion and interaction among different organs promises to be a useful approach. Recently, preliminary studies on 3D fluoroscopic treatment imaging and tumor localization based on principal component analysis motion models and cost function optimization have shown encouraging results. However, the performance of this technique for varying breathing parameters and under realistic conditions remains unclear and thus warrants further investigation. In this work, we present a systematic evaluation of a 3D fluoroscopic image generation algorithm via two different approaches. In the first approach, the model's accuracy is tested for changing parameters for sinusoidal breathing. These parameters include changing respiratory motion amplitude, period and baseline shift. The effects of setup error, imaging noise and different tumor sizes are also examined. In the second approach, we test the model for anthropomorphic images obtained from a modified XCAT phantom. This set of experiments is important as all the underlying breathing parameters are simultaneously tested, as in realistic clinical conditions. Based on our simulation results for more than 250 s of breathing data for eight different lung patients, the overall tumor localization accuracies of the model in left-right, anterior-posterior and superior-inferior directions are 0.1 +/- 0.1, 0.5 +/- 0.5 and 0.8 +/- 0.8 mm, respectively. 3D tumor centroid localization accuracy is 1.0 +/- 0.9 mm.

Digital phantoms continue to play a significant role in modeling and characterizing medical imaging. The currently available XCAT phantom incorporates both the flexibility of mathematical phantoms and the realistic nature of voxelized phantoms. This phantom generates images based on a regular breathing pattern and can include arbitrary lung tumor trajectories. In this work, we present an algorithm that modifies the current XCAT phantom to generate 4D imaging data based on irregular breathing. First, a parameter is added to the existing XCAT phantom to include any arbitrary tumor motion. This modification introduces the desired tumor motion but, comes at the cost of decoupled diaphragm, chest wall and lung motion. To remedy this problem diaphragm and chest wall motion is first modified based on initial tumor location and then input to the XCAT phantom. This generates a phantom with synchronized respiratory motion. Mapping of tumor motion trajectories to diaphragm and chest wall motion is done by adaptively calculating a scale factor based on tumor to lung contour distance. The distance is calculated by projecting the initial tumor location to lung edge contours characterized by quadratic polynomials. Data from ten patients were used to evaluate the accuracy between actual independent tumor location and the location obtained from the modified XCAT phantom. The RMSE and standard deviations for ten patients in x, y, and z directions are: (0.29 ± 0.04, 0.54 ± 0.17, and0.39 ± 0.06) mm. To demonstrate the utility of the phantom, we use the new phantom to simulate a 4DCT acquisition as well as a recently published method for phase sorting. The modified XCAT phantom can be used to generate more realistic imaging data for enhanced testing of algorithms for CT reconstruction, tumor tracking, and dose reconstruction.