The underlying image data that is used to characterize tumors is provided by medical scanning technology. Instead of taking a picture like a camera, the scans produce raw volumes of data which must be further processed to be usable in medical investigations. To get actual images that are interpretable, a reconstruction tool must be used.
There are a variety of reconstruction algorithms, so consideration must be taken to determine the most suitable one for each case, as the resultant images will differ. This influences the quality and usability of the images, which in turn determines how easily an abnormal finding can be detected and how well it can be characterized.
The reconstructed images are saved in a large database. A public database to which all clinics have access enables broadly collaborative and cumulative work in which all can benefit from growing amounts of data, ideally enabling a more precise workflow.
After the images have been saved in the database, they have to be reduced to the essential parts, in this case the tumors, which are called “volumes of interest”.
Because of the large image data that needs to be processed, it would be too much work to perform the segmentation manually for every single image if a radiomics database with lots of data is created. Instead of manual segmentation, an automated process has to be used. A possible solution are automatic and semiautomatic segmentation algorithms. Before it can be applied on a big scale an algorithm must score as high as possible in the following four tasks:
- First, it must be reproducible, which means that when it is used on the same data the outcome will not change.
- Another important factor is the consistency. The algorithm does solve the problem at hand and performs the task rather than doing something that is not important. In this case, it is necessary that the algorithm can detect the diseased part in all different scans.
- The algorithm also needs to be accurate. It is very important that the algorithm detects the diseased part in the most precise way possible. Only with accurate data, accurate results can be achieved.
- A minor but still important point is the time efficiency. The results should be generated as fast as possible so that the whole process of radiomics can also be accelerated. A minor point means in this case that, if it is in a certain frame, it is not as important as the others.
Features Extraction and Qualification
After the segmentation, many features can be extracted and the relative net change from longitudinal images (delta-radiomics) can be computed. Radiomic features can be divided into five groups: size and shape based–features, descriptors of the image intensity histogram, descriptors of the relationships between image voxels (e.g. gray-level co-occurrence matrix (GLCM), run length matrix (RLM), size zone matrix (SZM), and neighborhood gray tone difference matrix (NGTDM) derived textures, textures extracted from filtered images, and fractal features. The mathematical definitions of these features are independent of imaging modality and can be found in the literature. A detailed description of texture features for radiomics can be found in Parekh, et al.,(2016) and Depeursinge et al. (2017).
Due to its massive variety, feature reductions need to be implemented to eliminate redundant information. Hundreds of different features need to be evaluated with a selection algorithms to accelerate this process. Additionally, features that are unstable and non-reproducible should be eliminated since features with low-fidelity will likely lead to spurious findings and unrepeatable models.
After the selection of features that are important for our task it is crucial to analyze the chosen data. Before the actual analysis, the clinical and molecular (sometimes even the genetic) data needs to be integrated because it has a big impact on what can be deducted from the analysis. There are different methods to finally analyze the data. First, the different features are compared to one another to find out whether they have any information in common and to reveal what it means when they all occur at the same time.
Another way is Supervised or Unsupervised Analysis. Supervised Analysis uses an outcome variable to be able to create prediction models. Unsupervised Analysis summarizes the information we have and can be represented graphically. So that the conclusion of our results is clearly visible.
Several steps are necessary to create an integrated radiomics database. The imaging data needs to be exported from the clinics. This is already a very challenging step because the patient information is very sensitive and governed by Privacy laws, such as HIPAA. At the same time the exported data must not lose any of its integrity when compressed so that the database only incorporates data of the same quality. The integration of clinical and molecular data is important as well and a large image storage location is needed.
The goal of radiomics is to be able to use this database for new patients. This means that we need algorithms that run new input data through the database which return a result with information about what the course of the patients’ disease might look like. For example, how fast the tumor will grow or how good the chances are that the patient survives for a certain time, whether distant metastases are possible and where. This determines how the further treatment (like surgery, chemotherapy, radiotherapy or targeted drugs etc.) and the best solution which maximizes survival or improvement is selected. The algorithm has to recognize correlations between the images and the features, so that it is possible to extrapolate from the data base material to the input data.
Applications of Radiomics
Prediction of clinical outcomes
Aerts et al. (2014)performed the first large-scale radiomic study that included three lung and two head-and-neck cancer cohorts, consisting of over 1000 patients. They assessed the prognostic values of over 400 textural and shape- and intensity-based features extracted from the computed tomography (CT) images acquired before any treatment. Tumor volumes were defined either by expert radiation oncologists or using semiautomatic segmentation methods.Their results identified a subset of radiomic features that may be useful for predicting patient survival and describing intratumoural heterogeneity. They also confirmed that the prognostic ability of these radiomics features may be transferred from lung to head-and-neck cancer. However, Parmar et al. (2015)demonstrated that prognostic value of some radiomic features may be cancer type dependent. Particularly, they observed that not every radiomic feature that significantly predicted the survival of lung cancer patients could also predict the survival of head-and-neck cancer patients and vice versa.