Aims and objectives
FDG-PET/CT is the diagnostic standard procedure for staging of NSCLC patients.
The TNM staging classification contains both morphological and dimensional criteria for each stage,especially for T-stage.
The manual application of these criteria during PET/CT reading is time-consuming and error-prone.
Radiomics allows for the extraction of quantitative texture features that are not obvious for the human eye.
The study aims to show which of these radiomics-metrics may be used to automate T-staging.
Methods and materials
98 PET and CT image datasets from patients who underwent FDG-PET/CT for staging of NSCLC were extracted from the PACS and transferred to a 3D Slicer-based annotation software that allows for a manual tumor segmentation.
Tumor segmentation was performed by an experienced radiology and nuclear medicine physician.
For each patient a status of ‘diameter-based staging’ (n=50; DBS) or ‘morphological upstaging’ (n=48; MU) were assigned.
Next,
radiomics were compiled for each patient covering shape,
CT,
and PET textures (see bar plot for excerpt).
An ensemble of...
Results
The decision tree trained on a small-group of input data (70%) achieved an AUC score for predicting the tumors to upstage of 0.89.
Since decision trees were used,
the resulting analysis can be easily read out as simple true/false rules with the corresponding weights as probabilities.
Mean Absolute Deviation of the CT in Hounsfield units being less than 120 corresponds with an over 95% chance of being up-staged.
Conclusion
We could demonstrate that there is a wide array of radiomic features that allow a precise,
accurate subcategorization beyond the conventional diameter measurements without requiring manual examination of the tissue morphology.
ROC analysis indicates that these features are important candidates for both refinement of clinical staging and implementation of software algorithms that can automatically detect and stage NSCLC.
Given the large degree of uncertainty and difficulty in staging by standard guidelines,
the use of clearly defined radiomics metrics which can be automatically calculated could simplify...
References
Ribeiro,
M.
T.,
Singh,
S.
and Guestrin,
C.
(2016) ‘“Why Should I Trust You?” Explaining the Predictions of Any Classifier’.
doi: 10.1145/2939672.2939778.
Goo,
J.
M.
(2011) ‘A Computer-Aided Diagnosis for Evaluating Lung Nodules on Chest CT: The Current Status and Perspective’,
Korean Journal of Radiology,
12(2),
pp.
145–155.
Lee,
K.
H.
et al. (2012) ‘Computer-aided detection of malignant lung nodules on chest radiographs: Effect on Observers’ performance’,
Korean Journal of Radiology,
pp.
564–571.
Mankoff,
D.
A.,
Muzi,
M.
and Zaidiy,
H.
(no date) ‘Quantitative Analysis...