Bootstrap Framework 3.3.6

[jose_carlos_pichardo]

I have posted this question before but no one answered. In order to generate the DICOM reference surface from the CT SIM image set one has to specify a HU value threshold. Clearly the surface will be different depending on what HU threshold value one selects. Is there anyone out there who has investigated the effect of selecting different HU threshold values for the generation of the DICOM BODY/External surface used in AlignRT? How did you design your testing process to determine the optimum HU threshold value? Is it site-specific? Surely the selection of a HU threshold will affect the DIBH breath hold amplitude accuracy.

[sherri_thornton]

Our routine is to transfer the data set to Eclipse from the CT SIM. All contouring is done in Eclipse. We use the default HU setting for BODY always.

[marko_laaksomaa]

Hi Jose
If you use FB and BH- surfaces in DIBH setup which are generated at the CT (without daily the same couch value in vertical at the setup), HU- values should not have any effect on BHL amplitude itself, neither the BHL you see in AlignRT from FB to BH surface, nor BHL in the LAT image (vertebra vs. sternum). This is because I suppose the same named possible error in the BODY exists in both FB and BH reference surfaces if the BODY is created with the same HU- values in both surfaces. Then in the setup, from zero VRT error in FB to zero VRT error in BH, the BHL itself should be ok. For the sternum VRT accuracy, and thereby PTV VRT isocenter accuracy, this thing may have significance and is therefore important anyhow. Additionally, if in the DIBH, you use averaged (daily the same) couch value in the setup, and acquire FB reference surface during the first fractions like we do, also BHL with described unsuccesful BODY in BH DICOM is finally slightly different than planned. Then there is a need to acquire also new DIBH reference surface for the setup. With -350 HU, which is use in all the treatment sites here, I have noticed -1.5 mm group average error in the sternum VRT in DIBH DICOM, with +- 2 mm BHL window in VRT. Sternum remains too dorsally in the DIBH kV- LAT images. The case is the same in other sites also concerning bony structures. All the group averages in couch shifts in VRT in the IGRT studies are in minus, indicating the need to raise the couch to meet the DRR bony structures. This knowledge is not that significant in the treatments where we shift the couch VRT based on online image match daily, and daily acquire new reference surfaces for the treatment. As you mentioned in DIBH for example, this may have significance, if we daily setup the patient with such DICOM, which leads to errors in sternum VRT in the acquisition position and we do not correct the residual error with image guidance. Small errors are then possible in treatment, if we do not ask patient to take more or less air as much as the sternum VRT error is, or if we do not shift the couch in VRT, patient in BH. This small systematic residual error in sternum VRT is not necessarily visible in the tangential image ribs in VRT direction. To me it seems like we should create the BODY after scaling down the HU- values that much that the external surface in planning CT begins to be that white that it begins to pixelate and is not that accurate any more. Better solution would be to create external structure 1-2 mm outside the BODY. In the FB cases it should be taken into account that patients may relax right after setup and therefore that is the reason for need to raise the couch VRT based on image guidance, and it is possible that AlignRT DICOM surface finally agrees with the couch shifts in VRT. Therefore, if you want to evaluate this in practice with FB patients, it could be useful to check the AlignRT VRT in FB cases with DICOM after the couch shifts from the large number of patients, in the cases where the online image matching is done based on bony structures near the surface. If the AlignRT VRT is still on average +1.5 mm after the couch shifts is done in online match, before acquiring the new surface for the treatment, you have similar findings with me. KV-setup images and patients are not possible the best way to evaluate these things, because there may be for example pitch in the sternum and it causes uncertainties to select correct matching location, but if the offline data of hundreds of patients shows the same result, it improves the reliability. However, please test this on your own. We are talking about 1- max. 2 mm errors now, but certainly about systematic errors, which should be taken seriously if those exist in the group and user wants to go on with the DICOM without daily couch VRT corrections or even without IGRT, since possible swelling in the ROI area during the treatment may even worsen the accuracy systematically. Once we have the results of group average with SKIN, 1 or 2 mm outside the BODY, I will let you know.

BR Marko

[marko_laaksomaa]

Now we have 6 DIBH- patients with SKIN DICOM, created 2 mm outside BODY structure, with HU value -350. Based on the first fraction, 2/6 of the patients, BODY would be better in VRT, since sternum located 2 mm too ventrally in the LAT image with SKIN. For those two, new reference BH- surface was acquired. 3/6 of the patients had zero error in the sternum VRT on the first fraction with SKIN. 1/6 of the patients sternum located 1 mm too ventrally, indicating that the structure in between the BODY and SKIN would have been optimal. Group mean is +0.8 mm (sternum needs to be more dorsally), indicating that the 1 mm structure outside BODY would be optimal in general, not 2 mm outside BODY. For 4/6 of the patients DICOM was used during the whole course of treatment. If I take account all the treatment fractions from those patients and evaluate the residual error to the sternum in the images in the acquisition position (with SKIN DICOM), group mean error to the sternum in VRT is only 0.1 mm, indicating that finally SKIN (2 mm outside BODY) is optimal in the group. This is may be due to increased swelling in the ROI area during the whole treatment course. But, this finding will not entirely eliminate the need to acquire new setup surfaces, if we want to eliminate the small systematic errors individually. One possibility would be to setup with SKIN on the first fraction and on the second fraction switch to BODY, if it seems to work better concerning VRT isocenter in the images and if we want to stick with the DICOM and if we are interested in the 1-2 mm systematic errors in VRT. With the +-2 mm BH VRT threshold we need to pay good attention to check where the AlignRT VRT delta is exactly, at the time LAT image was acquired in the online match (patient in BH), if we want to compare that VRT delta reliably to the sternum VRT in the image acquisition position. This is just what I have found here in Tampere. There may be several things in the entire workflow, which affects on VRT accuracy and there is a possibility that this conclusion is not relevant in other hospitals and thereby this kind of finetuning will not work in other hospitals. This is why I suggest you to test these things on your own. On the other hand I suppose you have also found some kind challenges with this topic, since you ask this question.

[marko_laaksomaa]

Finally, I combine some results to my answer and think what this means from our side. I would like to say that I think it is worth somehow on the first fractions to achieve zero systematic error to the sternum VRT or tangential image ribs VRT in the images, as is your goal Jose. Based on the results, random error for the sternum VRT with SGRT only (without daily IGRT corrections) in the 20 patients DIBH group in our latest offline data is only 1.4 mm (group mean value of the patientspecific SDs for the daily sternum VRT in image acquisition position). With RPM here, with daily IGRT online match couch corrections, random error in the recent data (patient n=20) is 1.1 mm (p=0.01). Thereby daily 2D/2D IGRT and couch corrections based on online match still improves the accuracy also concerning this sternum VRT topic when compared to AlignRT only data (without image guidance). 1.4 mm SD in sternum VRT with AlignRT only (without IGRT corrections), seems however good enough for its part to proceed mostly with verification tangential images only, without online match couch corrections. As mentioned, this calls for that systematic error in sternum VRT or tangential image ribs VRT is in practice zero in the first fractions with the used setup reference surfaces without online match couch corrections. +- 2 mm BH window was used in both groups.

[marko_laaksomaa]

I did some additional calculations for this topic today, since I feel this interesting and important in practice. I calculated 48 DIBH- patients first fractions sternum VRT location in image acquisition position from the time BODY structure was used, with -350 HU value. The 48 different DIBH patients first fraction sternum VRT errors in LAT image, in the image acquisition position, with +- 2 mm AlignRT BH threshold in VRT were (mm): -2.4, -3.1, -1.9, -0.5, -2.2, -1, 0.9, -1.1, -1.4, 0, 3.2, -0.9, -4.1, 1.3, -5.4, 0, -4.6, -2.9, 6.1, 0, -0.3, -0.8, -0.5, -2.6, 0.5, 0.2, -1.3, 2.2, -1.1, 1.1, -4.2, -4.1, -0.8, -3.1, 1.1, -2.4, -1.3, 0.8, -2.2, -1.9, -3.1, -1.1, -1, -0.6, -4.2, -2.1, 4.7, -1.6. The group mean for those first fraction values is -1 mm. 71% of all the values are in minus, sternum in the images too dorsally compared to DRR. 23% of all the values are positive, sternum too ventrally. 40% of all the values are larger than 2 mm, indicating that DICOM needs to be replaced with new reference surface that often, due to unsuccessful VRT in the images, if the IGRT action level for sternum VRT error is 2mm. Corresponding percentage for 3 mm or larger sternum VRT displacement was 23% (8/11 of these values are minus). If I add in Excel 1 mm to those each individual values listed above (indicating 1 mm margin outside BODY= SKIN= BODY+1mm), 31% of all the values are still larger than 2 mm, even though the group mean is 0. 3 mm or larger sternum displacement exists with BODY+ 1 mm in 21% of the cases (6/10 of these values are minus). If the BH window is +-2 mm, that provides for its part an explanation for 0-2 mm errors in sternum VRT. With BODY+1mm sternum sternum VRT error sign is in minus in 52% of the cases and in plus in 44% of the cases. With this 1 mm margin outside the BODY, we can only slightly drop down the needs to acquire new setup reference surfaces for the setup. Our physicists noticed that there is a possibility to adjust DICOM isocenter in VRT,LNG and LAT in the AlignRT workstation offline, if we want to keep DICOM in the setup, in the cases where we notice small systematic isocenter errors in the images. Have you taken that into consideration? We have found it easier to take a new reference surfaces for the setup, at the time there is a need to correct systematic error. I have handled only VRT in my answer, since this question has most to do with the VRT I suppose. On the other hand, there is also different issues which leads us to give up DICOM surface, such as systematic errors in the images concerning rotations, individual structure such as shoulder joint etc. I think that there is not much possibility to know beforehand, which ones will be those “3 mm or larger cases or problematic cases” and forecast that with adjusting the HU- values. Isocenter/monthly calibration is one issue, which may shift this accuracy with 1 mm and some may think that this is our issue. It is difficult to believe that this is our issue, that long time I have had this mean error in the results. With BODY+1mm we can go on with DICOM in about 80% of the cases concerning sternum VRT in DIBH (with our 3 mm AL for sternum VRT) and I think that is good result, taking consideration how much can happen in between the CT and first treatment fraction in the patient anatomy for example. I hope there was something useful for you also.