Radiomics Guided Radiation Therapy – Applications And Emerging Opportunities
Radiomics in radirotherapy involves diagnosing, treating, and prognosis lung, esophageal, breast, rectal, and prostate cancers.
Radiation treatment aims to provide a high therapeutic dosage of radiation to malignancies while preserving healthy normal tissue. In radiotherapy, quantitative imaging has improved target volume definition, treatment response monitoring, and dose painting.
Radiomics is a cutting-edge image-processing paradigm that uses mathematical methods to extract a large variety of mineable quantitative characteristics. Radiomics-guided radiation therapy (RGRT) seeks to assist doctors and physicists in providing more effective radiation treatment.
Radiomics research investigations have revealed that radiomic characteristics may be utilized to enhance detection and diagnosis, prognosis, therapeutic response evaluation and prediction, and normal tissue damage.
Radiomics characteristics may be utilized to discover intra-tumor heterogeneities, which can subsequently be used to stratify patients and pick individually appropriate therapies.
Radiomics-based image analysis has been addressed using hand-crafted (explicit) features rather than deep learning-based features. Hand-created (HC) features are taken from a medical picture that has been segmented manually, semi-automatically, or completely automatically.
These characteristics are pre-defined and are utilized as input variables for outcome modeling or other research/clinical objectives following extraction and selection. On the other hand, deep learning-based features are created by a computer method and are not pre-defined.
Based on the established computational technique, these characteristics may be extracted from segmented and non-segmented pictures and directly employed for research/clinical purposes.
The basic categories of hand-crafted characteristics are shape subtypes, first-order, second-order, and high-order features. These characteristics are defined as follows, albeit these definitions may differ in the literature.
- Shape characteristics: Shape features are morphological descriptors of the image's segmented sections of tissues, such as characteristics quantifying the size and shape of a 2D and 3D segmentation's region of interest (ROI) and volume of interest (VOI). These characteristics include volume, surface area, sphericity, compactness, axis length, elongation, flatness, and diameter.
- Histogram-based characteristics: First-order features are those that are based on histograms. These characteristics are measurements of voxel/pixel intensity distribution inside the VOI/ROI. Percentiles (e.g., 10 and 90), energy, entropy, interquartile range, kurtosis, maximum, mean absolute deviation, mean, median, minimum coverage, robust mean absolute deviation, root mean squared, skewness, total energy, uniformity, and variance are all common histogram-based properties.
- Second-order or textural characteristics: Second-order features, also known as textural features, are notable features that assess the spatial distribution of the voxel/pixel intensity inside the VOI/ROI. Textural characteristics measure the statistical link between the intensity levels of surrounding pixels/voxels or groups of pixels/voxels as estimated by numerous descriptive matrices.
- Higher-order characteristics: Higher-order features are computed using particular mathematical transformations or filters on ROI/VOI, such as Wavelet, Laplacian of Gaussian, Fractal, and Fourier transform. These filters/transformations are very effective for highlighting specific photo patterns, such as edges and gradients.
Artificial neural networks, such as convolutional neural networks, calculate and identify deep learning-based (or deep) features (CNN). As a result, characteristics are created by the network rather than being pre-defined/hand-crafted.
Furthermore, the deep features obtained may be used with other machine learning and regression techniques. Deep learning-based radiomics is referred to as 'discovery radiomics' in certain publications.
Different designs, including stacks of linear and non-linear functions, may be used to extract deep learning characteristics. Deep learning-based radiomics into three categories: (1) an input hierarchy, (2) pre-trained and raw models, and (3) deep learning network designs.
The deep learning network may accept three levels of patient data for input: single slice, total volume, or complete patient examination. Depending on time and data quantity, several pre-training models may be trained or tweaked.
For example, although training a deep network from scratch has some advantages, it also has drawbacks, such as the curse of overfitting and class imbalance. However, there are alternatives, such as the use of transfer learning.
Finally, selecting the best deep network architecture is a critical challenge in deep learning-based radiomics. CNNs, Recurrent Neural Networks (RNNs), and Generative adversarial networks are some deep learning radiomics (GANs) models.
The primary characteristics for patient and treatment selection in radiation are tumor staging and grading. Radiomics factors may predict stage and grade in various malignancies, including prostate, lung, kidney, and colorectal cancer.
Another research employed CT radiomics characteristics combined with machine learning to predict pathologic stages of non-small cell lung cancer. Radiomics may give higher knowledge beyond conventional staging and grading by anticipating key subclinical illness states.
Some studies have used radiomics traits to predict human papillomavirus (HPV) status in the head, neck, and oropharyngeal malignancies.
Radiogenomics has the potential to become a speedier, non-invasive approach to analyzing the biological markers of each patient's cancer. Traditional biopsy and genetic analysis are time-consuming and intrusive for the patient to identify a patient's tumor characteristics.
Several cancer signature genes can be deciphered by imaging characteristics, according to radiogenomics research. In certain tumors, for example, radiomics signatures collected from CT, MRI, and positron emission tomography may adequately predict the mutation status of the cancer hallmark gene, epidermal growth factor receptor.
Radiomics examination of structural or functional pictures may aid in decoding regional radiosensitivity. Several aspects must be considered while choosing the best radiation treatment plan.
Radiomics research has shown that imaging characteristics may predict various clinical outcomes. Pre-treatment CT radiomics characteristics predicted overall survival (OS) in advanced lung adenocarcinoma.
Radiogenomics has shown that imaging characteristics have substantial relationships with genes implicated in radiation signaling pathways in several malignancies.
Imaging characteristics that correspond well with genetic data might be utilized to adjust radiation planning and delivery. A radiomics/radiogenomics sensitivity index might be used to create a new radiosensitivity index.
Several studies have used radiomics as a decision-making tool for patient/therapy selection. The European Society of Radiology approved using radiomics in clinical studies based on expert opinion.
The cancer kind dictates whether structural or functional pictures are collected for radiation patients and therapies. CT, positron emission tomography, and MRI are the most regularly utilized imaging modalities for target and organ at risk delineation.
Different imaging methods would be employed for patient-specific treatment planning in the age of precision radiotherapy. Volume definition or target localization is a critical component in radiation planning that assures treatment success.
The biological target volume (BTV) is a novel therapeutic volume that intends to improve therapeutics by incorporating molecular processes from positron emission tomography-CT imaging.
One proposed modification is the radiomics target volume (RTV), or radiomics area of interest; several ways may be used to compute sub-regions in RTV. A radiomics map is initially produced from the reference picture in one method.
This map is thought to be a novel picture that can be readily created by using radiomics feature equations. A radiomics volumetric signature map (RVSM) depicts the link between the radiation beam and tumor heterogeneities as determined by radiomics analysis.
Using radiomics feature algorithms on medical pictures, this map might be created. The radiation treatment plan might be based on the mapped heterogeneity.
A more sophisticated dosage delivery system might be 'Dose Painting by Feature' (DPBF). In this example, tumor targets depicted via radiomics mapping (RTVs) might be classified as high or low-risk based on feature values.
For example, low energy, high variance, low correlation, or common inverse difference moment areas might be used to predict irradiation tumor control using radiomics properties.
Dosimetric and geometric verification are required to ensure radiation therapy quality. There has been little research on radiomics as a strong and reliable method for controlling dosage delivery quality.
Radiomics characteristics were employed to improve the identification of IMRT faults in research by Wootton et al. Features extracted from dose maps might be utilized for dosimetric verification.
Adaptive radiotherapy, also known as treatment re-planning based on target temporal changes over a radiation course, is a method that may account for changes in patient weight, posture, or tumor volume.
Pre-post, during-post, and post-during radiation delta radiomics were estimated from pre-radiation, and weekly MRI changes for rectal cancer patients predicted complete tumor response to radiotherapy.
In another research, CT radiomics analysis was used to predict patient outcomes such as overall survival, distant metastases, and local recurrence in non-small cell lung cancer patients. According to this research, radiomics characteristics might be employed in the clinic as an adaptive procedure for patient re-planning in both targets and organs at risk.
According to several research, post-RT radiomics properties may be predictive biomarkers.
In recent research, imaging characteristics retrieved from MRI obtained four weeks after neoadjuvant chemoradiation treatment predicted response in patients with locally advanced rectal cancer.
The prediction power of post-treatment characteristics was calculated to be 0.70 in this research.
Although pre-treatment characteristics are often employed to develop predictive imaging biomarkers for radiation purposes, research has shown that post-treatment features give more therapeutic information and may be used to predict post-treatment occurrences.
Some research has indicated that post-RT radiomics properties may be employed as predictive model factors in radiotherapy-induced toxicities in the bladder and rectum.
MRI radiomics is used to differentiate radionecrosis from progression following stereotactic body irradiation for brain oligometastasis.
Textural features recovered from O-(2-[18F]fluoroethyl)-L-tyrosine (18F-FET) positron emission tomography may be utilized to distinguish radiation harm from brain metastases recurrence, according to Lohmann and colleagues.
An integrated model was also developed and validated for distinguishing tumor recurrence from radiation necrosis in glioma patients in another investigation.
CT radiomics can measure and distinguish between local recurrence and non-recurrence in lung cancer patients having stereotactic ablative radiation treatment.
There are specific issues with using radiomics characteristics, although they might be reliable biomarkers for response evaluation.
Radiomics is a multi-step procedure with specific problems that vary based on the imaging modality. These are examples of image capture, image reconstruction, image processing, image segmentation, feature definition/extraction, consistency of extracted features, data analysis, and model building.
Several suggestions are made to solve such issues. These are examples of this in assessing the reproducibility and repeatability of radiomics characteristics, employing consistent procedures in the radiomics process, adhering to established standardization rules, and using high sample numbers.
Although several research has substantially increased our knowledge of heterogeneity, it is still not completely understood. Furthermore, heterogeneity is distributed among individual cells and tumor regions.
As a result, radiomics characteristics may fail to effectively identify targets or incorrectly depict targets for optimum dosage since the imaging equipment's resolution may not be enough to provide pictures at the cellular level.
Hypoxia, for example, is a well-known cause of radiation response failure that may be visualized with positron emission tomography/CT, but existing imaging modalities do not assess it well.
To demonstrate this, consider that oxygen distribution at the cellular level is in the m scale, but imaging resolution is limited to the mm or cm scale. This problem confounds the radiomics properties used to quantify hypoxic zones.
Biological/histological validation of radiomics characteristics must be performed to solve this issue.
This problem reduces the usefulness of radiomics characteristics for measuring hypoxic environments. However, biological/histological confirmation of radiomics properties may aid in overcoming this limitation.
Cancer cells divide according to their cell cycle and undergo local microenvironmental alterations due to external influences such as medication and nutrition.
In regular radiation activities, planning pictures are taken one week before treatment commencement at a period unaffected by microenvironmental changes.
When radiomics are employed, this necessitates using newly collected pictures for radiation planning. Likewise, research has shown that radiation causes immunological, vascular, and stromal alterations in the tumor microenvironment, which may enhance radioresistance and tumor recurrence.
These factors also alter tumor heterogeneity areas, which must be re-quantified using radiomics characteristics during radiation, and new radiomics target volumes must be defined.
Radiomics characteristics are mathematical/statistical equations that are used to medical pictures to measure the inter-relationships between image pixels to capture intra-tumoral heterogeneity.
Although various radiogenomics studies have been carried out to discover correlations between imaging characteristics and genomics parameters and follow biological pathways in tumors using radiomics measurements, it is still unclear how these aspects link to the underlying biology.
On the other hand, there are uncertainties and unanswered concerns about how radiomics characteristics are affected by imaging modalities and imaging settings.
It is unclear, for example, how characteristics taken from a molecular/functional picture indicate physiological/molecular correlates or how features retrieved from an anatomical image link to the genes responsible for phenotypic variances.
This will be significant in adaptive radiotherapy, where the goal is to integrate structural and functional changes in the planning process using radiomics characteristics.
In any event, the total biology is likely more complicated than imaging parameters can fully decode. Future prospective investigations will uncover applicable/decoded biological pathways by applying more precise people to imaging characteristics.
The radiomics community faces significant hurdles in analyzing radioomics data for accurate clinical applications. Currently, the bulk of machine learning and deep learning methods utilized in radiomics analysis cannot be directly comprehended (black box).
Algorithms that capture biological, clinical, physical and imaging factors are urgently needed to allow knowledge-based radiomics planning in radiotherapy.
Meanwhile, infrastructure such as imaging biobanks and big sample size databases will aid in the creation and validation of sophisticated models.
Image-guided radiation therapy (IGRT) is a kind of radiation therapy in which imaging methods are used throughout each treatment session. Radiation treatment employs high-energy radiation beams to treat cancer and noncancerous tumors.
There are various varieties of brachytherapy, each distinguished by a unique means of delivering radiation to the body: interstitial brachytherapy, intracavitary brachytherapy, intraluminal radiation treatment, and radioactively tagged molecules administered intravenously.
At Genesis Healthcare, cutting-edge radiation therapy includes intensity-modulated radiation therapy (IMRT) to precisely match the greatest cancer cell load and image-guided radiation therapy (IGRT) to provide each treatment with a high degree of precision.
The merging of deep-learning technologies with high-quality imaging (and other clinical) data is intended to enable next-generation radiotherapy. Image-guided radiation therapy approaches such as MR-Linacs, cone beam computed tomography-based dose delivery,
Tomotherapy devices, and positron emission tomography-based radiotherapy dose delivery (RefleXion) will make it easier to accomplish deep learning-radiomics radiomics-guided radiation therapy. From patient/treatment selection through response evaluation, deep learning networks will play an important part in radiotherapy workflow.