Radiomics is an emerging field that combines the disciplines of radiology and omics to extract large amounts of quantitative features from medical images using data-characterisation algorithms. At the intersection of precision medicine and radiology, radiomics aims to uncover patterns within image data that can be correlated with genomic profiles, disease characteristics, or outcomes, enhancing diagnostic, prognostic, and predictive accuracy.
The foundation of radiomics is the assumption that biomedical images contain more information than what meets the eye. With advanced computational methods, it’s possible to extract and analyse a multitude of quantitative features that reflect the underlying pathophysiology of a disease, such as cancer. These features can be related to the shape, size, texture, and intensity of the image regions and are not typically appreciated by the human eye.
Process of Radiomics
Image Acquisition and Standardisation
Radiomics begins with high-quality image acquisition from modalities such as computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), or ultrasound. The success of radiomics depends heavily on standardised imaging protocols since variations can significantly affect the extracted features’ reliability.
Following acquisition, the region of interest (ROI) needs to be accurately segmented. This involves delineating the contours of the tumour or tissue, which can be performed manually, semi-automatically, or automatically. The precision of segmentation is crucial, as it directly influences the quality of the extracted data.
With the ROI defined, hundreds to thousands of features are extracted using specialised software. These features can be grouped into several categories:
- First-order statistics describe the distribution of voxel intensities within the ROI without concern for spatial relationships, such as mean, variance, skewness, and kurtosis.
- Second-order statistics or textural features consider the spatial arrangement of voxels. These include features from grey-level co-occurrence matrices (GLCM), grey-level run-length matrices (GLRLM), and grey-level size zone matrices (GLSZM).
- Higher-order statistics are derived from filters or mathematical models applied to the image, such as wavelet transforms, which can highlight patterns not visible in the original image.
- Shape-based features quantify the three-dimensional shape and size of the ROI.
Extracted features undergo statistical or machine learning-based analysis to identify patterns and associations with clinical outcomes, genomic data, or other relevant information. The overarching goal is to develop models that can predict a disease’s behaviour, such as response to treatment or likelihood of recurrence.
Any radiomic signature identified must be validated in separate, ideally prospective, cohorts to ensure its applicability to different populations and settings.
Applications of Radiomics
In oncology, radiomics has been at the forefront. It provides insights into tumour heterogeneity, which is not always possible with a biopsy. By analysing a whole tumour, radiomics can offer a non-invasive method to predict tumour grade, treatment response, and patient prognosis. For example, certain textural features from a lung cancer CT scan may predict which tumours are more likely to respond to chemotherapy.
In neurology, MRI-based radiomics can help characterise brain tumours, differentiate between types, and even between treatment effects such as pseudoprogression and true tumour progression. CT and MRI radiomics are being explored in stroke to predict outcomes and response to therapies.
Radiomics also finds applications in cardiology, potentially providing new biomarkers for cardiac diseases by assessing tissue characteristics in a detailed and quantifiable manner.
Radiogenomics is a subset of radiomics that focuses on correlating imaging features with gene expression patterns. This field could help in identifying genetic mutations based on imaging characteristics, which can be particularly useful for cancers where genetic testing is not always feasible.
Challenges and Future Directions
The lack of standardisation in imaging protocols and feature extraction methods is a significant hurdle in radiomics. Efforts such as the Imaging Biomarker Standardization Initiative aim to address this challenge.
Big Data and Machine Learning
Radiomics generates vast amounts of data, requiring robust statistical and machine learning tools for analysis. These tools must be capable of handling high-dimensional data while avoiding overfitting and ensuring model generalizability.
Integration with Clinical Practice
For radiomics to become part of routine clinical practice, the models developed must be integrated with electronic health records and clinical workflows. This integration requires user-friendly software and decision support systems.
Ethical and Legal Considerations
As with all data-intensive fields, radiomics raises ethical issues concerning patient privacy and data security. Additionally, there are questions about the legal implications of AI-driven diagnostics and their errors.
Future research is moving towards combining radiomics with other ‘omics’ data (like genomics, proteomics) to create multi-modal models that could offer a more comprehensive understanding of diseases.
Radiomics represents a paradigm shift in how medical images are used, from qualitative interpretation by radiologists to quantitative analysis that can drive personalised treatment plans. The potential of radiomics lies in its ability to provide non-invasive biomarkers that reflect the biological complexity of diseases, contributing to the broader goal of precision medicine. However, realising this potential requires overcoming significant challenges related to standardisation, analysis, and integration into healthcare systems. With ongoing research and technological advancements, radiomics could soon become essential in the personalised healthcare landscape.You Are Here: Home »