Standardisation in medical image processing
Standardisation is important in the development and use of imaging biomarkers for several reasons:
- Reproducibility: Standardisation ensures that imaging biomarkers are measured and calculated in a consistent and reproducible manner, which is necessary for their validity and reliability.
- Comparison: Standardisation allows for the comparison of imaging biomarkers across different studies and populations, which is necessary for the development of clinical guidelines and for the identification of trends and patterns.
- Integration: Standardisation enables the integration of imaging biomarkers with other types of data, such as clinical, genetic, or molecular data, which can improve the understanding of disease processes and facilitate personalized medicine.
- Quality control: Standardisation helps to ensure the quality and accuracy of imaging biomarkers, which is necessary for their clinical utility and for the trust of healthcare professionals and patients.
- Collaboration: Standardisation facilitates collaboration and data sharing between different research groups and institutions, which can accelerate the development and validation of imaging biomarkers.
The Image Biomarker Standardisation Initiative (IBSI) is an international collaboration aiming to standardise the process of extracting and reporting features from medical imaging modalities.
With IBSI Chapter 1 the following objectives were achieved: (a) feature nomenclature and definitions, (b) define a standardised and generalisable processing scheme for feature calculation from common medical imaging modalities, (c) provide datasets and reference values for software calibration and verification (d) provide reporting guidelines to help study reproducibility and validation as published in the first IBSI article in the Radiology journal.
In 2024 we have completed IBSI Chapter 2 which aimed at standardising several commonly used convolutional filters in the radiomics workflow: Mean, Laplacian-of-Gaussian, Laws Kernels, Gabor Kernels, Separable Wavelets, Nonseparable Wavelets, Riesz Transformations. This work is described in this arXiv publication and in the second IBSI article in the Radiology journal.
SPAARC compliance with IBSI
SPAARC includes a total of 165 radiomic features standardised following IBSI guidelines as reported in Table 1.
| Type | Family Name | Baseline Feature Number |
| Shape-based | Morphological | 23 |
| First-order | Intensity-Based Statistics Intensity Histograms Intensity-Volume Histogram | 18 23 7 |
| Texture | Grey Level Co-occurrence Matrix (GLCM) Grey Level Run Length Matrix (GLRLM) Grey Level Size Zone Matrix (GLSZM) Grey Level Distance Zone Matrix (GLDZM) Neighbourhood Grey Tone Difference Matrix (NGTDM) Neighbourhood Grey Level Dependence Matrix (NGLDM) | 25 16 16 16 5 16 |
Cardiff were one of 5 teams in the first IBSI article to implement over 95% of the possible number of features considered for benchmarking through phases 1 and 2. Cardiff results for the configuration tested in the IBSI as compared to final benchmarks can be downloaded from Cardiff University digital repository of research outputs.
To help provide a complete and reliable report on image processing and feature extraction, SPAARC implements the IBSI recommended nomenclature and abbreviations.