Skip to main content

Table 8 The summary table of literature focused on extracting reproducible features

From: Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling

Modality

Disease

Variability

Statistical indicators

Reproducibility summary of radiomics features

References

PET

Drenal gland carcinoma, Lung, Epiglottis, and esophagus cancer

Acquisition modes

Reconstruction parameters

\({\text{\% Diff}} = \frac{{100 \times \left( {{\text{X}} - {\text{X}}_{{{\text{mean}}}} } \right)}}{{{\text{X}}_{{{\text{mean}}}} }}\)

Entropy-first order, energy, maximal correlation coefficient, low gray level run emphasis

[95]

CT

fILD

Scanners

Reconstruction settings (reconstruction kernels, slice thicknesses)

ICC

Radiomics of fILD are highly repeatable for constant reconstruction parameters in a single scanner, intra- and inter-scanner reproducibility are severely impacted by alterations in slice thickness more than reconstruction kernel

[96]

CT

Lung, liver and kidney tumors

Segmentation variability

ICC

Reproducibility: shape features > first order features > GLCM

[98]

CT (Phantom)

Lung cancer

CT acquisition parameters

Scanners

CCC, AUC

Tumor-mass, sigmoid-offset-mean, gabor-energy

[99]

CT

Liver tumor

CT radiation dose

Reconstruction settings (reconstruction section thicknesses, reconstruction kernels, reconstruction algorithms)

Hierarchical clustering

Reproducibility: shape features (including the maximum axial diameter and volume) > other features

[100]

MRI

Cervical cancer

Scanners

Segmentation readers

ICC

Reproducibility: shape features > other features

[101]

MRI (phantom)

Tumor

Scanners

ICC, COV

Reproducibility: first-order features > other features

[102]

  1. fILD fibrosing interstitial lung disease, ICC intraclass correlation coefficients, CCC concordance correlation coefficient, AUC area under receiver operating characteristic curve, COV coefficient of variation, GLCM grey level co-occurrence matrix