prostate cancer breast cancer lung cancer5y
Experiment 18 Methods: A single patient was used for this analysis The radiotherapy planning structures had been drawn on the CT AVIP in Eclipse (Varian Medical Systems, Palo
Alto). Eclipse contains a tool, which can auto-generate a structure from one phase of the 4DCT to other phases. A lung contour ipsilateral to the tumour and contra-lateral to the tumour was created for all phases of a scan of a single patient. These images were extracted and each volume was run twice, firstly standardised to itself using individualised quantisation and secondly standardised to the quantisation levels for the GTV for this patient (i.e. uniform quantisation).
114 Experiment 18 Results and discussion:
Figure 12 shows the comparison all phases of the 4D CT for the contra-lateral lung. The most unexpected observation is that all phases except the AVIP have an obvious high density low entropy region in the contra-lateral lung. In the ipsilateral plots seen in figure 13, all plots including the AVIP have a high density low entropy region. The other observation is that for phases CT20-40 and CT60-90, they all have a split high density low entropy region. This is seen in both the plots containing tumour and the plots not containing tumour. This raises the suspicion that the double plot is an effect of movement.
Figure 12: Elephant plots from textural analysis of a whole lung contra-lateral to the tumour generated from all phases of a 4DCT. Analysis included 10 4DCT phases and 2 composite volumes for the contra-lung of a patient having a radiotherapy planning 4DCT for SABR. None of the analysed structures contained tumour.
The elephant plots seen in figure 12 did not have standardised axes. The highest density point in the structure defines the X axis, this changes the relative shape of the data plots to each other making it difficult to compare different data plots. The high density low entropy region is an artefact due to how the structures for the different phases were generated. Eclipse has an auto-contouring tool that propagated the structure from the AVIP to the other phases. In these phases the structures were not reproduced faithfully, they included high density pixels related to rib and chest wall.
115
Figure 13: Elephant plots from textural analysis of a whole lung ipsilateral to the tumour generated from all phases of a 4DCT. 10 phases and 2 composite volumes for the ipsilateral- lung of a patient having a radiotherapy planning 4DCT for SABR. All of the analysed structures contained tumour.
The region of interest tool was used to transpose the data points of selected plots back into the original texture map using the co-ordinates of the original image. The results are seen in figure 14. The same definition as experiment 17 to analyse the high density low entropy region, i.e. density score of 900-1200 and entropy score of 2 or less. Figure 14 shows that the region of analysis from the high density low entropy region consistently overlaps with the position of the tumour in the original texture map, irrespective of the phase of the 4DCT used in the analysis.
116
Figure 14: Overlap of region of interest from data plot (density score 900-1200 and entropy score of 2 or less) with texture map from analysis of whole lung ipsilateral to tumour. a)= CT0, b)=CT20, c)=CT50, d)=CTAVIP and e)=CTMIP
As part of this experiment, it was noted the contents of the analysed structures were similar. What was not clear was whether FiniteRT would set similar quantisation level boundaries for similar structures. Optimal quantisation sets the levels based on the data in the structure being analysed. The values of the quantisation levels were compared between the CT0 of ipsilateral lung and contra-lateral lung, as well as to the quantisation levels of the GTV. The results can be seen in figure 15. As the tumour makes up a small amount of the total lung, it was not clear whether the presence of tumour in the lung volume would affect quantisation.
117
Figure 15: morphology of the elephant plots and quantisation levels for similar Regions of Interest. a) + b) standardised to self = individualised quantisation using LMQ. c) standardised to GTV = uniform quantisation using quantisation levels from GTV texture analysis.
Figure 15 shows that the individualised quantisation levels for the lung containing tumour and not containing tumour are broadly similar, but are different to the uniform quantisation levels set by the GTV. As a result this variability suggests that the values of the quantisation levels should be fixed to compare different structures. As the aim is to identify the presence of tumour, it was felt it would be reasonable to use tumour quantisation values to pre-dispose the plot to be able to identify tumour rather than other structures.
118
5.1.8 Experiment 19: Does uniform quantisation of texture maps affect data plot morphology?
Aim: The aim was to understand the effects of using individualised and uniform quantisation levels, to see if standardising the quantisation of data allows comparison of different analysed volumes. This experiment used the data plots from experiment 18, to understand the effects of optimised quantisation vs uniform quantisation.
Experiment 19 Methods: This experiment was made up of 2 parts. Firstly repeating the