Limits...
Interactive Volumetry Of Liver Ablation Zones.

Egger J, Busse H, Brandmaier P, Seider D, Gawlitza M, Strocka S, Voglreiter P, Dokter M, Hofmann M, Kainz B, Hann A, Chen X, Alhonnoro T, Pollari M, Schmalstieg D, Moche M - Sci Rep (2015)

Bottom Line: For the quantitative and qualitative analysis of the algorithm's results, manual slice-by-slice segmentations produced by clinical experts have been used as the gold standard (which have also been compared among each other).The results show that the proposed tool provides lesion segmentation with sufficient accuracy much faster than manual segmentation.The visual feedback and interactivity make the proposed tool well suitable for the clinical workflow.

View Article: PubMed Central - PubMed

Affiliation: Department of Neuroscience and Biomedical Engineering, Aalto University, Rakentajanaukio 2 C, 02150 Espoo, Finland.

ABSTRACT
Percutaneous radiofrequency ablation (RFA) is a minimally invasive technique that destroys cancer cells by heat. The heat results from focusing energy in the radiofrequency spectrum through a needle. Amongst others, this can enable the treatment of patients who are not eligible for an open surgery. However, the possibility of recurrent liver cancer due to incomplete ablation of the tumor makes post-interventional monitoring via regular follow-up scans mandatory. These scans have to be carefully inspected for any conspicuousness. Within this study, the RF ablation zones from twelve post-interventional CT acquisitions have been segmented semi-automatically to support the visual inspection. An interactive, graph-based contouring approach, which prefers spherically shaped regions, has been applied. For the quantitative and qualitative analysis of the algorithm's results, manual slice-by-slice segmentations produced by clinical experts have been used as the gold standard (which have also been compared among each other). As evaluation metric for the statistical validation, the Dice Similarity Coefficient (DSC) has been calculated. The results show that the proposed tool provides lesion segmentation with sufficient accuracy much faster than manual segmentation. The visual feedback and interactivity make the proposed tool well suitable for the clinical workflow.

No MeSH data available.


Related in: MedlinePlus

GrowCut initialization for the segmentation of the RF ablation zone: the ablated zone is marked in green and the background is marked in yellow on three 2D slices, respectively.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4612735&req=5

f7: GrowCut initialization for the segmentation of the RF ablation zone: the ablated zone is marked in green and the background is marked in yellow on three 2D slices, respectively.

Mentions: For comparison of our employed method with a state-of-the-art segmentation approach we used GrowCut40, which is freely available under the medical platform 3D Slicer (Fig. 6). The implementation is very user friendly because it does not require any precise parameter setting, rather the user initializes the method by marking areas in the image with simple strokes (Fig. 7). Furthermore, we had good experiences with certain types of brain tumors (Glioblastoma multiforme (GBM)41 and pituitary adenomas42). However, we tested the GrowCut algorithm with our RFA datasets and especially the cases where the needles are still in place – and thus visible within the images – caused massive problems (Fig. 8). What happens is, that GrowCut leaks along the needles, because it cannot handle the large gray value differences between the ablation zone (dark) and the RFA needle (bright). Overall we tested four cases from our datasets with GrowCut: two cases with the needles still in place and two cases without needles. For the two cases with needle in place, we could only archive a DSC of 50.64% (for the case presented in Fig. 8) and a DSC of 50.28% for the second case. As mentioned before, for both cases GrowCut leaks along the RFA needles, which explains the low Dice Scores. However, for the cases without needle in place we could archive Dice Scores of 80.3% and 76.29%. Here, in contrast to the cases with needle, the leaking did not occur, which resulted in acceptable segmentation results. But still, our approach could achieve higher Dice Scores of 82.23% and 83.43%, respectively. Moreover, the initialization of GrowCut (marking parts of the lesion and the background) takes between 30 and 60 seconds for a trained user, in contrast to our method that needs only a single seed point. In addition, the user has to wait several seconds for the GrowCut segmentation result (on the same PC we measured about 10 seconds for the interactive method), whereas our method provides the segmentation result immediately in real-time. This makes a refinement much more convenient. Note: the sharp edges of the GrowCut result (green) in the rightmost image of Fig. 8 occur, because the Slicer implementation restricts the segmentation area with a bounding box. The size of the bounding box depends on the initialization of the user and avoid GrowCut to use the whole image or volume for the automatic segmentation process. In addition to GrowCut we also tested and evaluated existing (implemented) segmentation approaches from other medical platforms, like The Medical Imaging Interaction Toolkit (MITK, www.mitk.org) and MeVisLab (see Results section) with our data. MITK, developed by the German Cancer Research Center (DKFZ) in Heidelberg, Germany, combines also the Insight Toolkit (ITK, www.itk.org) and the Visualization Toolkit (VTK, www.vtk.org) with an application framework. Amongst others, we applied the Fast Marching 3D implementation from MITK (version MITK-2015.05) to our data, but it could also not handle the extreme gray value differences between the needle (bright) and the ablation zone (dark). However, in contrast to GrowCut the Fast Marching algorithm did not leak along the needle, rather it excluded the needle. Also placing additional seed points direct on the needle and thus providing the algorithm the information about the bright parts did not lead to a satisficing segmentation result. The DSCs for the two cases with the needles still in place were 51.94% and 30.01%. But for the none needle cases we could achieve better results, which resulted in Dice Scores of 73.56% and 63.93% compared with the manual segmentations. However, beside the manual seed points that had to be placed for the approach to run, there are several segmentation parameters (Sigma, Alpha, Beta, Stopping value and Threshold), which make it on one hand quite difficult to find a good parameter setting. On the other hand the segmentation results could probably be better with a more precise parameter tuning. We also tried the Region Growing 3D from MITK but for all cases (needle and none needle) the approach leaked into the surrounding structure. However, if the seed point was placed directly on the needle within the image, the whole needle could be segmented quite well (due to the bright values of the needle). As Level-Set segmentation method we used the itkGeodesicActiveContourLevelSetImageFilter-module from the current MeVisLab version (MeVisLab 2.7, 18–06–2015), which wraps the ITK class GeodesicActiveContourLevelSetImageFilter43. Beside an arbitrary amount of user-defined seed points several parameters can be tuned for the segmentation, but a parameter change and re-segmentation took over two minutes, which made it quite time-consuming to find a good setting (note: the same laptop – described in the result section and used for the presented interactive segmentation – was also used for the MITK and MeVisLab segmentation approaches). However, the first case (with needle) showed a DSC of 60.98% and leaked along the needle. The second case with needle had only a DSC of 22.56% and leaked along the needle and into the surrounding structures. This happened because the gray values of the ablation zone were more similar to the surrounding tissue for this case. For the none needle cases we achieved Dice Scores of 45.63% and 70.18% and the lower DSC resulted, because we could only archive an under or an over segmentation of the ablation zone (the DSC presented here is for the under segmentation).


Interactive Volumetry Of Liver Ablation Zones.

Egger J, Busse H, Brandmaier P, Seider D, Gawlitza M, Strocka S, Voglreiter P, Dokter M, Hofmann M, Kainz B, Hann A, Chen X, Alhonnoro T, Pollari M, Schmalstieg D, Moche M - Sci Rep (2015)

GrowCut initialization for the segmentation of the RF ablation zone: the ablated zone is marked in green and the background is marked in yellow on three 2D slices, respectively.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4612735&req=5

f7: GrowCut initialization for the segmentation of the RF ablation zone: the ablated zone is marked in green and the background is marked in yellow on three 2D slices, respectively.
Mentions: For comparison of our employed method with a state-of-the-art segmentation approach we used GrowCut40, which is freely available under the medical platform 3D Slicer (Fig. 6). The implementation is very user friendly because it does not require any precise parameter setting, rather the user initializes the method by marking areas in the image with simple strokes (Fig. 7). Furthermore, we had good experiences with certain types of brain tumors (Glioblastoma multiforme (GBM)41 and pituitary adenomas42). However, we tested the GrowCut algorithm with our RFA datasets and especially the cases where the needles are still in place – and thus visible within the images – caused massive problems (Fig. 8). What happens is, that GrowCut leaks along the needles, because it cannot handle the large gray value differences between the ablation zone (dark) and the RFA needle (bright). Overall we tested four cases from our datasets with GrowCut: two cases with the needles still in place and two cases without needles. For the two cases with needle in place, we could only archive a DSC of 50.64% (for the case presented in Fig. 8) and a DSC of 50.28% for the second case. As mentioned before, for both cases GrowCut leaks along the RFA needles, which explains the low Dice Scores. However, for the cases without needle in place we could archive Dice Scores of 80.3% and 76.29%. Here, in contrast to the cases with needle, the leaking did not occur, which resulted in acceptable segmentation results. But still, our approach could achieve higher Dice Scores of 82.23% and 83.43%, respectively. Moreover, the initialization of GrowCut (marking parts of the lesion and the background) takes between 30 and 60 seconds for a trained user, in contrast to our method that needs only a single seed point. In addition, the user has to wait several seconds for the GrowCut segmentation result (on the same PC we measured about 10 seconds for the interactive method), whereas our method provides the segmentation result immediately in real-time. This makes a refinement much more convenient. Note: the sharp edges of the GrowCut result (green) in the rightmost image of Fig. 8 occur, because the Slicer implementation restricts the segmentation area with a bounding box. The size of the bounding box depends on the initialization of the user and avoid GrowCut to use the whole image or volume for the automatic segmentation process. In addition to GrowCut we also tested and evaluated existing (implemented) segmentation approaches from other medical platforms, like The Medical Imaging Interaction Toolkit (MITK, www.mitk.org) and MeVisLab (see Results section) with our data. MITK, developed by the German Cancer Research Center (DKFZ) in Heidelberg, Germany, combines also the Insight Toolkit (ITK, www.itk.org) and the Visualization Toolkit (VTK, www.vtk.org) with an application framework. Amongst others, we applied the Fast Marching 3D implementation from MITK (version MITK-2015.05) to our data, but it could also not handle the extreme gray value differences between the needle (bright) and the ablation zone (dark). However, in contrast to GrowCut the Fast Marching algorithm did not leak along the needle, rather it excluded the needle. Also placing additional seed points direct on the needle and thus providing the algorithm the information about the bright parts did not lead to a satisficing segmentation result. The DSCs for the two cases with the needles still in place were 51.94% and 30.01%. But for the none needle cases we could achieve better results, which resulted in Dice Scores of 73.56% and 63.93% compared with the manual segmentations. However, beside the manual seed points that had to be placed for the approach to run, there are several segmentation parameters (Sigma, Alpha, Beta, Stopping value and Threshold), which make it on one hand quite difficult to find a good parameter setting. On the other hand the segmentation results could probably be better with a more precise parameter tuning. We also tried the Region Growing 3D from MITK but for all cases (needle and none needle) the approach leaked into the surrounding structure. However, if the seed point was placed directly on the needle within the image, the whole needle could be segmented quite well (due to the bright values of the needle). As Level-Set segmentation method we used the itkGeodesicActiveContourLevelSetImageFilter-module from the current MeVisLab version (MeVisLab 2.7, 18–06–2015), which wraps the ITK class GeodesicActiveContourLevelSetImageFilter43. Beside an arbitrary amount of user-defined seed points several parameters can be tuned for the segmentation, but a parameter change and re-segmentation took over two minutes, which made it quite time-consuming to find a good setting (note: the same laptop – described in the result section and used for the presented interactive segmentation – was also used for the MITK and MeVisLab segmentation approaches). However, the first case (with needle) showed a DSC of 60.98% and leaked along the needle. The second case with needle had only a DSC of 22.56% and leaked along the needle and into the surrounding structures. This happened because the gray values of the ablation zone were more similar to the surrounding tissue for this case. For the none needle cases we achieved Dice Scores of 45.63% and 70.18% and the lower DSC resulted, because we could only archive an under or an over segmentation of the ablation zone (the DSC presented here is for the under segmentation).

Bottom Line: For the quantitative and qualitative analysis of the algorithm's results, manual slice-by-slice segmentations produced by clinical experts have been used as the gold standard (which have also been compared among each other).The results show that the proposed tool provides lesion segmentation with sufficient accuracy much faster than manual segmentation.The visual feedback and interactivity make the proposed tool well suitable for the clinical workflow.

View Article: PubMed Central - PubMed

Affiliation: Department of Neuroscience and Biomedical Engineering, Aalto University, Rakentajanaukio 2 C, 02150 Espoo, Finland.

ABSTRACT
Percutaneous radiofrequency ablation (RFA) is a minimally invasive technique that destroys cancer cells by heat. The heat results from focusing energy in the radiofrequency spectrum through a needle. Amongst others, this can enable the treatment of patients who are not eligible for an open surgery. However, the possibility of recurrent liver cancer due to incomplete ablation of the tumor makes post-interventional monitoring via regular follow-up scans mandatory. These scans have to be carefully inspected for any conspicuousness. Within this study, the RF ablation zones from twelve post-interventional CT acquisitions have been segmented semi-automatically to support the visual inspection. An interactive, graph-based contouring approach, which prefers spherically shaped regions, has been applied. For the quantitative and qualitative analysis of the algorithm's results, manual slice-by-slice segmentations produced by clinical experts have been used as the gold standard (which have also been compared among each other). As evaluation metric for the statistical validation, the Dice Similarity Coefficient (DSC) has been calculated. The results show that the proposed tool provides lesion segmentation with sufficient accuracy much faster than manual segmentation. The visual feedback and interactivity make the proposed tool well suitable for the clinical workflow.

No MeSH data available.


Related in: MedlinePlus