Download PDF

 

Radiomics is a rapidly evolving field in medical imaging, offering significant potential to improve clinical decision-making. This methodology extracts detailed quantitative features from medical images, revealing information that is imperceptible to the naked eye. However, the complexity of radiomics introduces challenges related to reproducibility, generalisability and clinical implementation. To address these, the European Society of Medical Imaging Informatics (EuSoMII) has established recommendations to guide researchers in radiomics practices.

 

Feature Extraction and Standardisation

The extraction of radiomics features begins with standardising images to ensure consistency across varying imaging equipment and conditions. Image pre-processing involves methods such as intensity normalisation, resampling and application of image filters, which are crucial for achieving reliable radiomics outcomes. For instance, intensity normalisation corrects variations in voxel intensity, which is essential when working with MRI, PET or CT images. EuSoMII highlights the need for researchers to adhere to guidelines such as those provided by the Image Biomarker Standardisation Initiative (IBSI) to enhance the reliability of extracted features.

 

The initiative recommends convolutional filters that are modality-independent when applying filters to derive higher-order features. By standardising the filters and extraction parameters, researchers can reduce variability and ensure the robustness of their models. Furthermore, comprehensive reporting of all image pre-processing steps is advised to maintain transparency and reproducibility.

 

Best Practices in Radiomics Model Development

Radiomics model development involves several critical steps, each susceptible to errors that can compromise the accuracy and clinical relevance of the model. Data partitioning is a fundamental task where incorrect handling can lead to bias and overfitting. To prevent this, it is essential to partition data at the patient level and employ resampling methods like cross-validation. This ensures that the training and validation processes are separated, avoiding information leakage between the datasets.

 

Additionally, selecting outcome parameters systematically is vital to ensure the clinical applicability of the developed models. The Radiology AI Deployment and Assessment Rubric (RADAR) framework offers a structured approach to select relevant parameters, particularly when moving from technical to clinical validation phases. Researchers should also employ multiple statistical tests when comparing models to fully understand the implications of their performance.

 

Increasing Clinical Translation through Explainability

The clinical adoption of radiomics depends not only on the accuracy of the developed models but also on their interpretability. EuSoMII's recommendations emphasise the importance of explaining how features correlate with clinical outcomes. To achieve this, methods like SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) are employed. These techniques provide insights into feature contributions and help clinicians understand the implications of the radiomics findings.

 

Explainability also involves using radiomics-specific guidelines such as the CheckList for EvaluAtion of Radiomics research (CLEAR) and the METhodological RadiomICs Score (METRICS). These guidelines ensure a systematic approach to feature selection, thereby enhancing model interpretability. Radiomics research must align with these guidelines to bridge the gap between advanced computational techniques and clinical practice.

 

Radiomics presents a transformative approach to medical imaging by offering more profound insights into diagnostic images. However, the methodological complexity poses challenges to its clinical adoption. Researchers can overcome these barriers by standardising feature extraction methods, adhering to robust model development practices and prioritising model explainability. Following EuSoMII’s recommendations will improve the reproducibility and generalisability of radiomics models, ultimately enabling their successful integration into clinical workflows. Adopting these best practices can unlock the full potential of radiomics, advancing patient care and decision-making in medical imaging.

 

Source: European Radiology

Image Credit: iStock

 


References:

Santinha J, Pinto dos Santos D, Laqua F et al. (2024) ESR Essentials: radiomics—practice recommendations by the European Society of Medical Imaging Informatics. Eur Radiol: In press.de


radiomics, medical imaging , EuSoMII, clinical decision-making, reproducibility, feature extraction, image standardisation, model development, explainability, radiomics models, clinical adoption

Latest Articles

Radiomics, medical imaging, clinical decision-making, feature extraction, EuSoMII, image standardisation, model development, explainability, reproducibility, radiomics models, clinical adoption Discover how EuSoMII’s guidelines in radiomics improve clinical decision-making through robust model development and feature standardisation.