Introduction:
Medical imaging techniques can provide information on tumors and its surrounding. Characteristics of tumors in sagittal and temporal planes can be visualized. The application of artificial intelligence in radiology has converted the obtained radiographic images to mineable data that are needed for oncology care. This process was first described in 2012 as radiomics. Since then, it has been increasingly used in the medical field to obtain information about medical images.
What Is Radiomics?
Radiomics is a field of medicine in which a large number of quantitative features are extracted from medical images using data characterization algorithms. The data enables the doctors to better treatment planning. Diseases that are difficult to visualize with the human eye can be discovered through radiomics. In recent years there has been methodological development in radiomics, including data acquisition, tumor segmentation, feature extraction, and modeling, and the inclusion of deep learning technologies.
Through radiomics, basic features like shape and size to advanced texture analysis metrics are extracted. Investigations are ongoing for the utilization of radiomics for tumor staging, treatment response, and patient outcome prediction. Initially, in radiomics, high-dimensional quantitative features were extracted in large amounts from multi-modality radiographic images like computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and ultrasonography. These images are correlated for determining features that are useful for diagnosis or predicting the prognosis of cancer. Overall, information that is available from clinical charts, treatment responses, genomic assays, and radiomics is helpful for predicting the outcomes of cancer.
Radiomics can help make decisions on treatment and outcomes by providing quantitative and objective support. This is accomplished by obtaining data from radiography images and combining them with clinical and other information, and mining data to identify radiomic biomarkers.
What Is the Technological and Methodological Development in Radiomics?
Radiomics is a multi-step process and includes data acquisition, preprocessing, tumor segmentation, feature extraction, knowledge discovery, and modeling.
1. Data Acquisition and Preprocessing
The first step involved in radiomics is data acquisition. Medical images taken from CT, MRI, PET, or ultrasonography are collected from various data centers or hospitals. Parameters and protocols are established to collect images, and software applications are utilized for their reconstruction.
For CT images, differences in scanners, voxel size (pixel), and reconstruction kernel can cause discrepancies in radiomics. Recent studies have shown that the use of a controlled protocol can reduce variation in radiomic features. In PET imaging, large variations can occur due to changes in grid size, reconstruction algorithm, and the number of iterations (repetition of operation). Respiratory motion can also create significant variations. Therefore, the use of respiratory-gated 4D (dimensional) PET scans can reduce motion blurring and generate robust images. MRI studies were conducted on the effectiveness of field strength, imaging protocols, and manufacturing differences for obtaining radiomic features. Significant discrepancies were not found between images. Ultrasound imaging is useful for differentiating between benign and malignant tumors in thyroid and breast cancers. A standard protocol for acquisition and image normalization can enable the reproducibility of images.
2. Tumor Segmentation
In radiomics, the entire primary tumor is analyzed. Tumor segmentation is essential to plan the region that requires further analysis and is an important step. Segmentation of tumors can be achieved through manual, automatic, or semiautomatic methods. Manual methods can affect image reproducibility and stability. Hence, automatic and semiautomatic methods are preferable.
3. Feature Extraction
Images with a high amount of features must be extracted. The features obtained are of two types manually defined or deep learning features. Manually defined features are of semantic and non-semantic types depending on the possibility of mathematical expression. Deep learning features are derived from algorithms of neural networks. Data from deep learning features are more specific to clinical outcomes.
4. Knowledge Discovery and Modeling
A large number of quantitative features obtained from tumors can make information redundant and overfitting. Hence, features must be selected to obtain relevant information.
The selected features are constructed into a prediction model to determine clinical outcomes. Machine learning can provide various modeling methods. The generated model must be validated to determine the potential for clinical application. Externally validated models are considered more credible. There are several tools available for the evaluation of models.
What Is the Application of Radiomics in Determining Treatment Outcomes for Cancer?
-
Brain Tumors: Primary and secondary malignant tumors of the brain are evaluated with CT, MRI, or PET imaging. In patients with glioblastoma, radiomics was used to predict various responses to treatment. This could significantly cut the cost of treatment planning.
-
Head and Neck Cancer: Radiomics is widely used for this type of cancer. In advanced nasopharyngeal cancers, MRI-based radiomics can be used as a prognostic factor. Tumor failure or occurrence of xerostomia following radiation therapy can be determined with radiomics.
-
Breast Cancer: It is the most common cancer in women worldwide. Multi-modality imaging data from radiomics and clinical information are being used for breast cancer research. The pathological complete response to neoadjuvant therapy and the prognosis of breast cancer can be predicted. MRI features and radiomics are combined to estimate disease-free survival in breast cancer patients.
-
Lung Cancer: It is a severe cancer with increasing incidence worldwide. Radiomics is used to diagnose, treatment planning, and prognosis for patients. In addition, radiomics is useful in predicting recurrence after treatment, stage III cancer response to radiotherapy, and response to neoadjuvant chemoradiation. Radiomics could predict prognosis from CT images.
-
Colorectal Cancer: This is a fatal type of cancer and requires chemoradiotherapy before surgery. The response to treatment can only be assessed following surgery. Therefore, radiomics can predict treatment response and eliminate unnecessary surgery. Individualized treatment strategies can be adopted based on radiomic to achieve improved treatment outcomes.
-
Prostate Cancer: Cancer is more prevalent in men. Radiomics can be used for the evaluation of treatment response. Data from MRI imaging can enable radiomics to predict recurrence following radical prostatectomy or radiotherapy.
-
Liver and Gastric Cancer: Radiomics plays a major role in their management. Overall survival and recurrence rates can be effectively analyzed with radiomics. Gastric cancer response to neoadjuvant therapy can be determined. With CT images prognosis for gastric cancer can be predicted.
What Are the Challenges for Radiomics?
Data is distributed in various hospitals, and their collection is time-consuming. Data sharing between countries is important, but it may lead to logistic problems. Reproduction and quality control are necessary for radiomics. Improving image quality can aid research purposes. Interpretability of radiomic features and models can vary based on different observers.
Conclusion:
Radiomics is a new field that is increasingly gaining interest in clinical application. There are rapid technological advancements in the field through the integration of deep learning techniques. The integration of oncology, radiology, and deep learning methods has gained importance for diagnosing, treatment planning, and predicting outcomes for cancer patients. Further studies on interpretability and reproducibility techniques are necessary to gain wider acceptance.
