Portal de Periódicos da CAPES

2797

2006; Elsevier BV; Volume: 66; Issue: 3 Linguagem: Inglês

10.1016/j.ijrobp.2006.07.1215

1879-355X

M Ding, Lei Xing, W Xiong, K. Stuhr, F. Newman,

Radiomics and Machine Learning in Medical Imaging

Purpose/Objective(s)Four-dimensional (4D) radiotherapy, the explicit inclusion of the temporal changes in anatomy during the imaging, planning and delivery of radiotherapy, has been investigated recently. We developed a 4D Monte Carlo treatment planning tool using an image interpolation model. In this study, we validated this 4D treatment planning tool with both a moving phantom and a real patient’s 4D data.Materials/MethodsThe 4D Monte Carlo treatment planning tool uses two initial 3D CT images at given breathing phases, and the 3D CT images at other phases are interpolated from these two given 3D CT data according to the patient’s breathing pattern. The correlation between the voxels in a reference CT image (one of the initial CT images) and the voxels in the CT images at other breathing phases was quantified using our linear interpolation model. The 4D dose was built up by mapping each 3D dose from all the CT images to the reference CT image based on the voxel correlation derived for the CT image interpolation. The Monte Carlo EGS4/MCDOSE code is used in the dose calculation, and the patient DICOM CT data are converted into relative CT image text files for Monte Carlo dose calculation using the ImageMagick software. A moving phantom and a patient’s 4D CT images were acquired by using a GE LightSpeed-QX/I scanner. The CT image interpolation model was validated by comparing the interpolated images to the CT images at different breathing phases for both 4D CT data. The final 4D dose (3D dose of the phantom CT images at 10 breathing phases mapped to the reference phantom CT image at the end inspiration) was compared to the measured dose of the moving phantom (the moving pattern during the beam delivery having been kept the same as that during the CT image acquisition).ResultsComparing the interpolated CT images to the phantom CT images at different phases, we found that the difference of the outlines of the outer contour and also the inner organs was within 2mm in all the directions. This difference is about the half the dimension of a voxel (3.5×3.5×3mm3 in this study), which is the minimum element shift in the model. Comparing the interpolated CT images to the patient’s 4D CT, while the difference of the outer contour was < 2mm, the difference of some inter organ positions was >2 mm in some directions. The bigger difference of the inter organ position was mainly due to the organ’s non-linear movement, which was not corrected in our model, and was not reflected by the breathing pattern we monitored. The difference of 90% and 100% isodose line of the prescribed dose (2Gy) was within 2mm, comparing the moving phantom measurement to the 4D dose calculation. The difference of the average of five point dose was within 2% comparing the measurement to the calculation.ConclusionsWe have validated our 4D treatment planning tool. The correlation between the voxels in the interpolated CT image and those in the reference CT image enables us to perform 4D planning even without 4D CT data. The clinical applications of our method in the 4D radiotherapy could reduce radiation dose and decrease the disk storage usage in 4D imaging. Purpose/Objective(s)Four-dimensional (4D) radiotherapy, the explicit inclusion of the temporal changes in anatomy during the imaging, planning and delivery of radiotherapy, has been investigated recently. We developed a 4D Monte Carlo treatment planning tool using an image interpolation model. In this study, we validated this 4D treatment planning tool with both a moving phantom and a real patient’s 4D data. Four-dimensional (4D) radiotherapy, the explicit inclusion of the temporal changes in anatomy during the imaging, planning and delivery of radiotherapy, has been investigated recently. We developed a 4D Monte Carlo treatment planning tool using an image interpolation model. In this study, we validated this 4D treatment planning tool with both a moving phantom and a real patient’s 4D data. Materials/MethodsThe 4D Monte Carlo treatment planning tool uses two initial 3D CT images at given breathing phases, and the 3D CT images at other phases are interpolated from these two given 3D CT data according to the patient’s breathing pattern. The correlation between the voxels in a reference CT image (one of the initial CT images) and the voxels in the CT images at other breathing phases was quantified using our linear interpolation model. The 4D dose was built up by mapping each 3D dose from all the CT images to the reference CT image based on the voxel correlation derived for the CT image interpolation. The Monte Carlo EGS4/MCDOSE code is used in the dose calculation, and the patient DICOM CT data are converted into relative CT image text files for Monte Carlo dose calculation using the ImageMagick software. A moving phantom and a patient’s 4D CT images were acquired by using a GE LightSpeed-QX/I scanner. The CT image interpolation model was validated by comparing the interpolated images to the CT images at different breathing phases for both 4D CT data. The final 4D dose (3D dose of the phantom CT images at 10 breathing phases mapped to the reference phantom CT image at the end inspiration) was compared to the measured dose of the moving phantom (the moving pattern during the beam delivery having been kept the same as that during the CT image acquisition). The 4D Monte Carlo treatment planning tool uses two initial 3D CT images at given breathing phases, and the 3D CT images at other phases are interpolated from these two given 3D CT data according to the patient’s breathing pattern. The correlation between the voxels in a reference CT image (one of the initial CT images) and the voxels in the CT images at other breathing phases was quantified using our linear interpolation model. The 4D dose was built up by mapping each 3D dose from all the CT images to the reference CT image based on the voxel correlation derived for the CT image interpolation. The Monte Carlo EGS4/MCDOSE code is used in the dose calculation, and the patient DICOM CT data are converted into relative CT image text files for Monte Carlo dose calculation using the ImageMagick software. A moving phantom and a patient’s 4D CT images were acquired by using a GE LightSpeed-QX/I scanner. The CT image interpolation model was validated by comparing the interpolated images to the CT images at different breathing phases for both 4D CT data. The final 4D dose (3D dose of the phantom CT images at 10 breathing phases mapped to the reference phantom CT image at the end inspiration) was compared to the measured dose of the moving phantom (the moving pattern during the beam delivery having been kept the same as that during the CT image acquisition). ResultsComparing the interpolated CT images to the phantom CT images at different phases, we found that the difference of the outlines of the outer contour and also the inner organs was within 2mm in all the directions. This difference is about the half the dimension of a voxel (3.5×3.5×3mm3 in this study), which is the minimum element shift in the model. Comparing the interpolated CT images to the patient’s 4D CT, while the difference of the outer contour was < 2mm, the difference of some inter organ positions was >2 mm in some directions. The bigger difference of the inter organ position was mainly due to the organ’s non-linear movement, which was not corrected in our model, and was not reflected by the breathing pattern we monitored. The difference of 90% and 100% isodose line of the prescribed dose (2Gy) was within 2mm, comparing the moving phantom measurement to the 4D dose calculation. The difference of the average of five point dose was within 2% comparing the measurement to the calculation. Comparing the interpolated CT images to the phantom CT images at different phases, we found that the difference of the outlines of the outer contour and also the inner organs was within 2mm in all the directions. This difference is about the half the dimension of a voxel (3.5×3.5×3mm3 in this study), which is the minimum element shift in the model. Comparing the interpolated CT images to the patient’s 4D CT, while the difference of the outer contour was < 2mm, the difference of some inter organ positions was >2 mm in some directions. The bigger difference of the inter organ position was mainly due to the organ’s non-linear movement, which was not corrected in our model, and was not reflected by the breathing pattern we monitored. The difference of 90% and 100% isodose line of the prescribed dose (2Gy) was within 2mm, comparing the moving phantom measurement to the 4D dose calculation. The difference of the average of five point dose was within 2% comparing the measurement to the calculation. ConclusionsWe have validated our 4D treatment planning tool. The correlation between the voxels in the interpolated CT image and those in the reference CT image enables us to perform 4D planning even without 4D CT data. The clinical applications of our method in the 4D radiotherapy could reduce radiation dose and decrease the disk storage usage in 4D imaging. We have validated our 4D treatment planning tool. The correlation between the voxels in the interpolated CT image and those in the reference CT image enables us to perform 4D planning even without 4D CT data. The clinical applications of our method in the 4D radiotherapy could reduce radiation dose and decrease the disk storage usage in 4D imaging.