Keywords:
Hybrid Imaging, Lung, Computer applications, PET-CT, Image manipulation / Reconstruction, Neural networks, Computer Applications-Detection, diagnosis, Outcomes analysis, Experimental investigations, Tissue characterisation, Cancer, Outcomes
Authors:
M. Kirienko1, L. Lozza2, N. Gennaro1, A. Rossi1, E. Voulaz1, A. Chiti1, M. Sollini1; 1Milan/IT, 2Bergamo/IT
DOI:
10.1594/ecr2018/C-2980
Results
The highest AUC obtained on the validation set was 0.79.
The corresponding model was a forest with 10000 trees,
6 as split dimension,
0.25/0.75 as relative weight and on the dataset containing CT and clinical features.
The dataset was the one with only the feature with importance greater than the 80th quantile.
In particular the nine variable se-lected were: ’Contrast_NGLDM’,
‘Correlation_GLCM',
‘EntropyH',
‘LZHGE_GLZLM',
‘SZHGE_GLZLM',
‘Sphericity',
‘maxValue',
‘minValue',
‘stdValue'.
The importance grade of the feature is illustrated in Figure 2.
The graphical representation of patients belonging to the training set,
depending on the nine features,
is illustrated in Figure 3.
A model with the same parameter but on the data with CT and clinical features,
with “gnum” and “Stage” and without “Age” gives a similar performance: 0.78 AUC.
Consider only the feature with importance greater than the 80th quantile result in the same of the first model but with “Stage” and without “EntropyH”.
Consequently the stage seems significant but not fundamental to predict the DFS.
It is observed that the clinical features (except “Stage”) are not relevant in general.
Indeed when they are in a dataset combined with imaging features,
their importance is under the 25th quantile.
Thereafter the variables “Age” and “gnum” are considered irrelevant.
Furthermore the performances on dataset with CT are always better than dataset containing only PET features.
For example the model with the same parameter but on the dataset with CT,
Clinical and PET features has 0.65 as AUC.
The variables considered in this case are 18 and contain all of the 9 features of the first model except “maxValue”.