Various modalities of images in clinics
The demand for the study of organismal physiology has led to an accelerating era of increased use of imaging from 1990 to the present, driven not only by additional data on structures and functions provided by imaging, but also by declining instrument and acquisition costs.
In cancer in particular, the development of imaging-guided radiotherapy has brought with it the routine acquisition of diagnostic and treatment planning images for patients, creating large datasets over the past decade that were until recently unexplored.
Imaging can generally be divided into two categories: non-invasive imaging and invasive imaging with varying degrees of the latter. Invasive methods involve some risk to the patient, including X-ray, fluoroscopy, computed tomography (CT), and positive emission tomography (PET).
Usually, the risk is limited to the radiation of the imaging device but can also be in the form of radiotracers injected into the patient to enhance contrast, or as in angiography with the risk associated with inserting a catheter into an artery in the groin that can cause death.
Non-invasive imaging has little to no risk, including MRI, electrocardiogram (ECG), and ultrasound. The distinction can also become ambiguous, for example in intravascular ultrasonography (IVUS). In IVUS, an acoustic transducer is passed into the heart through a catheter and is therefore invasive whereas the imaging itself, ultrasound, is not.
Along these same lines, imaging can be broadly categorized as structural and functional imaging. Structural imaging deals with the geometry and topology of the patient, whereas functional imaging is often temporal and focused on blood flow, oxygen flow, and anatomical function. Functional imaging tends to have low spatial resolution, at the expense of temporal or functional accuracy.
Structural imaging includes X-rays, CT, and MRI, while functional imaging includes functional magnetic resonance imaging (fMRI), ultrasound, positive emission emission imaging (PET), and electrocardiography (ECG). It should be noted, however, that this classification is general rather than binary, and therefore ultrasound may be used for example to measure prostate size in a single case, or as a more functionally directed part of the ECG.
In conclusion, imaging requires careful consideration of the additional benefit to the patient’s health course, the potential risk or harm to the patient, and its cost. As a result, the largest datasets used for the study are those related to routine care, where data are available for all patients on a particular care pathway.
The rise of radiomics: combine medical images with omics
The amount of medical imaging data has been recognized as an important task for computer-assisted decision support for approximately half a century; It has even been suggested that by 2020 there will be a “recursive dialogue” between doctors and computers during diagnostic and predictive tasks. Furthermore, different methods of data processing, feature extraction, and pattern recognition for classification of radiographs are described.
To date, there has been an abundance of research in the field of medical image processing and analysis, exploring everything from computer-assisted diagnosis to image-based markers for estimating neurological disease progression to the explosion of deep learning-based approaches.
More recently, quantitative features extracted from imaging data have shown correlation with tumor grade, histology, and treatment response; But successful integration into clinics has not yet been achieved. Only recently have increased computer power and the increased collection of electronic records and digital images made these technologies possible. In 2012, the automated process of quantifying image features was renamed “radiomics”.
Radiomics is experiencing increasing interest from researchers and clinicians. This interest is driven by developments in pattern recognition, computer vision, and model building that make the use of radiomics in diagnosis, prediction, and treatment decision-making processes more promising than ever before.
Radiomics has the potential to quantify information about the entire tumor and its different patterns within the tumor, making it a possible way to measure disease diversity in a non-invasive way that can be used in conjunction with interventional bio-histology and traditional quantitative imaging methods.
Studies have found predictors of recurrence after radiotherapy in lung cancer, predictors for patients receiving combination therapy with chemotherapy, radiotherapy, and targeted therapeutic agents, signatures of tumor metabolism and predictors of freedom from distant metastasis, locoregional control, and survival. The largest study to date was published in 2014 by Aerts et al, in which the performance of environmental radiomics of four prognostic features was investigated in lung and head and neck patients.
These signatures were subsequently shown to be a surrogate for tumor size, a known clinical prognostic factor. This has prompted interest in determining the strength and sustainability of radiomics features, and measures have been proposed to limit misinterpretation of results, increase reproducibility, and prevent this field from being classified as “another science.”