|Year : 2021 | Volume
| Issue : 3 | Page : 37-41
Application and prospect of radiomics in spinal cord and spine system diseases: A narrative review
Chao Ma, Guihuai Wang
Department of Neurosurgery, Beijing Tsinghua Changgung Hospital, School of Clinical Medicine, Tsinghua University, Beijing, China
|Date of Submission||22-Sep-2021|
|Date of Decision||11-Oct-2021|
|Date of Acceptance||15-Oct-2021|
|Date of Web Publication||11-Nov-2021|
Prof. Guihuai Wang
Department of Neurosurgery, Beijing Tsinghua Changgung Hospital, School of Clinical Medicine, Tsinghua University, Beijing
Source of Support: None, Conflict of Interest: None
Spinal cord and spine system diseases are complex and diverse, and prognosis is often poor. Therefore, early diagnosis is essential, especially for spinal system tumors, which are malignant nervous system tumors that have the highest mortality and disability rates. Accurate diagnosis avoids unnecessary operations. Traditional medical imaging diagnosis remains at the level of anatomical morphology, and there is a considerable amount of useful information that can be extracted and utilized. Radiomics is a new method of medical imaging diagnosis that is committed to improving image analysis and is capable of extracting a large number (more than 200 types) of quantitative features from medical images. Numerous studies on the application of radiomics in various systems of the body have been conducted. We reviewed current research on radiomics in spinal cord and spine system diseases and discussed the progress and challenges to provide a basis for improving the diagnosis and identification of spinal cord and spine system diseases and offer evidence-based support for precision medicine.
Keywords: Glioma, radiomics, review, spinal cord, spine
|How to cite this article:|
Ma C, Wang G. Application and prospect of radiomics in spinal cord and spine system diseases: A narrative review. Glioma 2021;4:37-41
|How to cite this URL:|
Ma C, Wang G. Application and prospect of radiomics in spinal cord and spine system diseases: A narrative review. Glioma [serial online] 2021 [cited 2023 Feb 3];4:37-41. Available from: http://www.jglioma.com/text.asp?2021/4/3/37/330195
| Introduction|| |
Medical imaging plays an indispensable role in disease diagnosis, staging and grading, treatment, and prognostic judgment. Radiomics is a high-throughput data analysis method that allows the acquisition of standard and quantitative data from medical images, which are then used to build models. It is accurate, objective, and repeatable, and multiple imaging methods can be combined to assist in medical imaging in clinical practice. Such noninvasive examination approaches have broad application prospects in the diagnosis and treatment of spinal diseases.
Radiomics is a noninvasive image analysis method based on machine learning and statistics to build models and provide clinicians with detailed, accurate, quantifiable, and reproducible indicators. In 2012, Lambin et al. were inspired by radio-genomics and developed the concept of radiomics, which was defined as, "using high-throughput technology extract imaging features from radiological images and create a usable database." Neurology has continuously developed since its origins in the nineteenth century, and our initial knowledge stemmed from studies on anatomy and biopsy., However, our current understanding of neurological diseases remains limited, and appropriate examination methods are lacking. With the development of imaging and the general trend of medical-industry integration, noninvasive diagnosis and treatment models based on radiomics have become the focus of neurological researchers. This article introduces the workflow of radiomics and reviews recent advances in clinical applications of radiomics to spinal cord neurological diseases. This article reviews the latest research progress in the applications of radiomics to spinal neurological diseases.
| Database Search Strategy|| |
The authors used the following inclusion criteria: Studies that discussed the application of radiomics to spine and spinal cord system diseases. English language and full-text articles published between January 2017 and June 2021 were included in this nonsystematic review. Participants were males and females of any age with a clinical diagnosis of spine or spinal cord diseases. The authors searched the PubMed database to identify relevant publications. The literature search strategy was conducted as follows: Each of three synonymous phrases, i.e., (1) spine, (2) spinal cord, and (3) glioma, were combined with radiomics. The authors screened the reference list of included studies to identify other potentially useful studies. First, the authors screened the titles and abstracts, then, the full texts for keywords, such as "radiomics" to find those that were potentially suitable.
| Radiomics Workflow|| |
Unlike traditional imaging, radiomics integrates various imaging methods, such as computed tomography, magnetic resonance imaging (MRI), positron emission tomography, and ultrasound, to link multidisciplinary knowledge and technology. The process primarily involves the acquisition of images, segmentation of images, extraction and screening of features, modeling, and analysis. Radiomics is based on extensive data analysis, and image acquisition equipment and parameters impact the features extracted by radiomics; therefore, standardized imaging solutions should be used to minimize unnecessary confounding factors. After obtaining a large amount of standardized medical imaging data, the accuracy of image segmentation, namely, the segmentation of target tissues, also affects radiomics analysis results. The operator can use manual, semiautomatic, or fully automatic image segmentation software, which differ in accuracy, resulting in deviations in the selection of radiomics features. In practice, the appropriate segmentation method should be selected according to the characteristics of different lesions.
Feature extraction and screening are performed using computer vision to convert images into quantifiable data and to classify and analyze the data. Standard radiomics features include morphological features as well as first-, second-, and high-order statistical outputs. Depending on the selected features, deviations or overfitting of the final results may occur, which can affect the sensitivity and specificity of the results. Modeling and analysis are the final steps of the radiomics analysis process. Statistical methods, machine learning, and artificial intelligence are combined with clinical data to assist in clinical diagnoses, differential diagnoses, treatment decision-making, and prognostic observation.
| Application of Radiomics to Spinal and Spine System Diseases|| |
Radiomics for the differential diagnosis of spinal cord metastases
Spinal metastases are the most common skeletal system tumors, and the spine is the most easily affected area. Bone metastasis can occur in almost all malignant tumors, of which the most common are breast, lung, and prostate cancer. Radiomics can help differentiate and diagnose various spinal cord and nervous system tumors, determine the stage of tumors, and stratify patients to identify the best treatment plan for each patient. Filograna et al. evaluated eight tumor patients (three lung cancer; one prostate cancer; one esophageal cancer; one nasopharyngeal cancer; one liver cancer; one breast cancer) with a total of 58 vertebral bodies (29 metastatic tumors and 29 nonmetastatic tumors) using T1-and T2-weighted images for lesion delineation and feature extraction. Using a Wilcoxon test, 16 significantly different features were detected between the metastatic and nonmetastatic groups, and the area under curve (AUC) values of the T1 and T2 radiomics models were 0.8141 and 0.916, respectively, which demonstrated that MRI-based radiomics analyses can distinguish between metastatic and nonmetastatic vertebral body tumors. Lang et al. analyzed 61 patients with spinal metastases (30 lung and 31 non-lung cancers) using a (Dynamic contrast enhancement MRI [DCE-MRI]) sequence to differentiate spinal metastases originating from primary lung cancer from other cancers. The convolutional neural network used for classification yielded an average accuracy of 0.71, whereas the convolutional long-and short-term memory network provided a higher accuracy of 0.81. These results showed that machine learning analysis methods based on radiomics features of DCE-MRI images have the potential to distinguish spinal lung cancer metastases. Zhong et al. developed and validated an MRI-based radiomics classification model to distinguish between cervical spine radionecrosis and metastasis in patients with nasopharyngeal carcinoma after radiotherapy. Contrast-enhanced T1-weighted images of 95 cervical spine lesions were acquired after radiotherapy in 46 patients with nasopharyngeal carcinoma to perform radiomics feature extraction, and a nomogram was developed using the least absolute shrinkage and selection operator algorithm. Results demonstrated that MRI-based imaging can be used as a noninvasive tool to distinguish cervical spine osteoradionecrosis from vertebral body metastasis after radiotherapy for nasopharyngeal carcinoma.
Radiomics studies of multiple sclerosis and optic neuromyelitis
Accurate diagnosis and early and appropriate treatment are essential to reduce the incidence of neuromyelitis optica and multiple sclerosis. Distinguishing optic neuromyelitis from multiple sclerosis based on clinical manifestations and neuroimaging remains challenging. Liu et al. retrospectively studied 67 patients with multiple sclerosis and 68 patients with neuromyelitis optica with spinal cord injury as the primary cohort and prospectively included 28 patients with multiple sclerosis and 26 neuromyelitis optica as the verification cohort. Nine radiological features extracted from spinal cord lesions were combined with clinical information, such as lesion length, sex, and Extended Disability Status Scale score, to develop an accurate predictive model to distinguish multiple sclerosis patients from neuromyelitis optica patients. The study by Ma et al. included 88 patients and 62 patients as the primary and verification cohorts, respectively, and applied a similar method. Quantitative radiological features from T2-weighted images were automatically extracted from the lesion area using the least absolute shrinkage and selection operator algorithm, and 11 radiomics features were selected. These radiomics features were combined with clinical features to build a nomogram model, which was found to accurately identify patients with neuromyelitis optica from those with multiple sclerosis.
Radiomics to detect osteoporosis based on lumbar magnetic resonance imaging
He et al. analyzed 109 cases of lumbar vertebral body T1- and T2-weighted MRI scans using dual-energy X-ray bone mineral density detection as a classification index to classify patients with osteoporosis. Their imaging bone loose classification model showed great potential for lumbar T1- and T2-weighted radiomics analysis in spinal bone detection.
| Prospects of Radiomics in Glioma of the Spinal Cord|| |
Automatic segmentation of spinal glioma
Glioma of the spinal cord is a relatively common tumor in the spinal canal. Its incidence is second only to schwannoma and meningioma and accounts for 15% of spinal canal tumors. Spinal glioma originates from glial and ependymal cells and grows in an infiltrating manner to form extensive dendritic tissues that infiltrate the surrounding tissues. There is no obvious boundary between the tumor and healthy tissue because of the heterogeneity of the tumor. Because of this, a simple threshold method cannot be used for segmentation. Glioma of the spinal cord is similar to brain gliomas. It is a typical tumor that contains multiple subdomains, such as active cells, necrosis, and edema, and often exhibits diverse visual characteristics, which include different shapes and appearances (e.g., necrotic and edema areas); moreover, the boundaries between different regions of the tumor are irregular. Studies have repeatedly demonstrated that glioma segmentation is a multiclassification problem and have emphasized the segmentation of tumor shape and specific tissues, such as active tumors and edema areas. Zikic et al. and Konukoglu et al. carried out classification studies on three types of gliomas and used random forest as a multiclass classifier to determine the type of tissue of each pixel. In the field of spinal glioma, automatic segmentation of spinal gliomas using radiomics and machine learning has also been shown to be feasible and promising.
Classification and grade of spinal glioma
Radiomics is valuable for diagnosing and classifying gliomas, guiding treatment, and predicting prognoses. MRI is the preferred approach for diagnosing spinal cord tumors. It has distinct advantages in soft tissue imaging and accurately displays the location and nature of the tumor. At present, clinical diagnoses are often assisted by artificial discrimination images. The World Health Organization guidelines for diagnosing and treating glioma divide glioma into four levels and provide corresponding diagnoses and treatment plans. Accurate grading is crucial for determining appropriate treatment plans, evaluating treatment response, and predicting patients' prognoses. Moreover, radiomics has significant clinical value in predicting the pathological type of gliomas before surgery. Qin et al. reported that in 66 cases of glioma, 114 radiomics features were extracted from preoperative MRI images (T2-weighted, T1-enhanced, and diffusion-weighted imaging). Pathological results were used as the gold standard to establish a predictive model of tumor grade before surgery, and the AUC of the receiver operating characteristic for the model was 0.943. This indicated that the model could accurately distinguish between high and low-grade glioma before surgery. It is believed that the same method may also be used in the classification of spinal glioma.
Radiomics predicts gene expression of spinal glioma
In addition to grading, gene expression using radiomics analysis is also a primary research focus. Revealing tumor gene expression status using image characteristics improves the prediction outcomes of patients with brain tumors and other diseases, assists in the clinical diagnosis of central nervous system diseases, and provides a noninvasive method for genotype detection. Numerous radiomics studies have been conducted in the field of brain glioma: Li et al. applied radiomics technology to the identification of MRI characteristics of low-grade gliomas and epidermal growth factor receptor using the MRI images of 270 patients with low-grade glioma. A total of 431 imaging features were extracted to examine the relationship between low-grade glioma and radiomics features, and logistic regression was used to select features related to epidermal growth factor receptor expression. Data were divided into a training (n = 200) and validation set (n = 70), which yielded an AUC of 0.95 and a classification accuracy rate of 90%. In another study, Li et al. investigated the p53 mutation status of low-grade gliomas based on the MRI images of 272 patients and obtained AUCs of 0.896 and 0.763 for the training and validation sets, respectively. In 117 MRI images, Li et al. used radiomics technology to predict K-67 expression in low-grade gliomas and achieved an accuracy rate of 88.6% in the validation set. Furthermore, Li et al. also used radiomics to predict the α-thalassemia/mental retardation syndrome X-linked mutation status of low-grade gliomas using the Cancer Genome Atlas database as the training and internal validation sets and the Chinese Glioma Genome Atlas as the external independent validation set. AUCs were 0.940, 0.925, and 0.725, respectively, which indicated that in low-grade gliomas, the radiological characteristics of T2-weighted images are related to α-thalassemia/mental retardation syndrome X-linked mutation status. Overall, the generalization performance of radiomics models is good. Thus, quantitative radiomics analysis may offer a noninvasive method for genotyping. Although numerous studies have been conducted on gene expression in spinal glioma, there remains limited research on spinal glioma using radiomics and gene expression. Nevertheless, future advances and integration between medicine and engineering will facilitate further research in this field.
Radiomics predicts spinal cord tumor treatment response
The study by Xi et al. demonstrated that radiomics can noninvasively predict the methylation status of the O6-methylguanine DNA methyltransferase promoter in patients with glioblastoma before surgery. The methylation status of the O6-methylguanine DNA methyltransferase gene promoter can be used as a basis for diagnosis, treatment, and prognostic evaluation, as well as for the prediction of response to temozolomide. Identifying patients who are not sensitive to temozolomide avoids the risk of overtreatment and the possible toxic side effects caused by chemotherapy. In addition, radiomics has been shown to predict the status of isocitrate dehydrogenase 1, which is highly correlated with the occurrence, treatment, and prognosis of glioma. Moreover, recent research has demonstrated that the integration of imaging omics, vital genetic molecules, and clinical features into a multilayered decision-making framework improves the accuracy of disease stratification and may promote personalized treatment of patients with glioblastoma.
| Limitations of Radiomics|| |
At present, the application of radiomics to the spine and spinal system is broadening; however, numerous challenges remain: (1) Radiomics research is predominantly single-center research, which differs from metabonomics, pharmacy/pharmacology, and surgery research, and the combination of disciplines is limited; (2) there are considerable differences in equipment, imaging parameters, and scanning schemes, analysis software and research methods are not standardized. Moreover, sample sizes are relatively small, which significantly limit the reproducibility and accuracy of results; (3) radiomics research is still in its infancy, and investigation of central nervous system diseases is relatively limited globally; (4) although there have been advances in the application of radiomics and genomics to the central nervous system, the mechanism underlying the quantification of the pathological process of tumors using radiomics characteristics remains unclear and requires further investigation.
| Summary and Outlook|| |
Radiomics enables reductions in medical costs; because most patients with spinal cord diseases have undergone imaging, the cost of radiomics analysis is relatively low. In addition, radiomics is a noninvasive method of examination, in contrast to biopsy, which is invasive, costly, and high risk. Finally, the early identification of patients who are unlikely to respond to chemotherapy using radiomics avoids unnecessary treatments and toxicity risks.
A fundamental component of radiomics is big data. The use of standardized data and radiomics will significantly promote the development of individualized medicine. Radiomics technology based on massive data samples, combined with the advances in deep learning technology, will help to facilitate personalized and precision medicine.,,
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG, Granton P, et al.
Radiomics: Extracting more information from medical images using advanced feature analysis. Eur J Cancer 2012;48:441-6.
Kumar V, Gu Y, Basu S, Berglund A, Eschrich SA, Schabath MB, et al.
Radiomics: The process and the challenges. Magn Reson Imaging 2012;30:1234-48.
Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al.
Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 2014;5:4006.
Glasser MF, Smith SM, Marcus DS, Andersson JL, Auerbach EJ, Behrens TE, et al.
The human connectome project's neuroimaging approach. Nat Neurosci 2016;19:1175-87.
Zhou M, Scott J, Chaudhury B, Hall L, Goldgof D, Yeom KW, et al.
Radiomics in brain tumor: Image assessment, quantitative feature descriptors, and machine-learning approaches. AJNR Am J Neuroradiol 2018;39:208-16.
Lambin P, Leijenaar RT, Deist TM, Peerlings J, de Jong EE, van Timmeren J, et al.
Radiomics: The bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017;14:749-62.
Polan DF, Brady SL, Kaufman RA. Tissue segmentation of computed tomography images using a random forest algorithm: A feasibility study. Phys Med Biol 2016;61:6553-69.
Incoronato M, Aiello M, Infante T, Cavaliere C, Grimaldi AM, Mirabelli P, et al.
Radiogenomic analysis of oncological data: A technical survey. Int J Mol Sci 2017;18:805.
Wong DA, Fornasier VL, MacNab I. Spinal metastases: The obvious, the occult, and the impostors. Spine (Phila Pa 1976) 1990;15:1-4.
Filograna L, Lenkowicz J, Cellini F, Dinapoli N, Manfrida S, Magarelli N, et al.
Identification of the most significant magnetic resonance imaging (MRI) radiomic features in oncological patients with vertebral bone marrow metastatic disease: A feasibility study. Radiol Med 2019;124:50-7.
Lang N, Zhang Y, Zhang E, Zhang J, Chow D, Chang P, et al.
Differentiation of spinal metastases originated from lung and other cancers using radiomics and deep learning based on DCE-MRI. Magn Reson Imaging 2019;64:4-12.
Zhong X, Li L, Jiang H, Yin J, Lu B, Han W, et al.
Cervical spine osteoradionecrosis or bone metastasis after radiotherapy for nasopharyngeal carcinoma? The MRI-based radiomics for characterization. BMC Med Imaging 2020;20:104.
Liu Y, Dong D, Zhang L, Zang Y, Duan Y, Qiu X, et al.
Radiomics in multiple sclerosis and neuromyelitis optica spectrum disorder. Eur Radiol 2019;29:4670-7.
Ma X, Zhang L, Huang D, Lyu J, Fang M, Hu J, et al.
Quantitative radiomic biomarkers for discrimination between neuromyelitis optica spectrum disorder and multiple sclerosis. J Magn Reson Imaging 2019;49:1113-21.
He L, Liu Z, Liu C, Gao Z, Ren Q, Lei L, et al.
Radiomics based on lumbar spine magnetic resonance imaging to detect osteoporosis. Acad Radiol 2021;28:e165-71.
Gooya A, Pohl KM, Bilello M, Cirillo L, Biros G, Melhem ER, et al.
GLISTR: Glioma image segmentation and registration. IEEE Trans Med Imaging 2012;31:1941-54.
Zikic D, Glocker B, Konukoglu E, Criminisi A, Demiralp C, Shotton J, et al.
Decision forests for tissue-specific segmentation of high-grade gliomas in multi-channel MR. Med Image Comput Comput Assist Interv 2012;15:369-76.
Konukoglu E, Glocker B, Zikic D, Criminisi A. Neighbourhood approximation forests. Med Image Comput Comput Assist Interv 2012;15:75-82.
Louis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK, et al.
The 2016 World Health Organization classification of tumors of the central nervous system: A summary. Acta Neuropathol 2016;131:803-20.
Qin JB, Liu Z, Zhang H, Shen C, Wang XC, Tan Y, et al.
Grading of gliomas by using radiomic features on multiple magnetic resonance imaging (MRI) sequences. Med Sci Monit 2017;23:2168-78.
Li Y, Liu X, Xu K, Qian Z, Wang K, Fan X, et al.
MRI features can predict EGFR expression in lower grade gliomas: A voxel-based radiomic analysis. Eur Radiol 2018;28:356-62.
Li Y, Qian Z, Xu K, Wang K, Fan X, Li S, et al.
MRI features predict p53 status in lower-grade gliomas via a machine-learning approach. Neuroimage Clin 2018;17:306-11.
Li Y, Qian Z, Xu K, Wang K, Fan X, Li S, et al.
Radiomic features predict Ki-67 expression level and survival in lower grade gliomas. J Neurooncol 2017;135:317-24.
Li Y, Liu X, Qian Z, Sun Z, Xu K, Wang K, et al.
Genotype prediction of ATRX mutation in lower-grade gliomas using an MRI radiomics signature. Eur Radiol 2018;28:2960-8.
Xi YB, Guo F, Xu ZL, Li C, Wei W, Tian P, et al.
Radiomics signature: A potential biomarker for the prediction of MGMT promoter methylation in glioblastoma. J Magn Reson Imaging 2018;47:1380-7.
Yu J, Shi Z, Lian Y, Li Z, Liu T, Gao Y, et al.
Noninvasive IDH1 mutation estimation based on a quantitative radiomics approach for grade II glioma. Eur Radiol 2017;27:3509-22.
Kickingereder P, Neuberger U, Bonekamp D, Piechotta PL, Götz M, Wick A, et al.
Radiomic subtyping improves disease stratification beyond key molecular, clinical, and standard imaging characteristics in patients with glioblastoma. Neuro Oncol 2018;20:848-57.
Parmar C, Grossmann P, Rietveld D, Rietbergen MM, Lambin P, Aerts HJ. Radiomic machine-learning classifiers for prognostic biomarkers of head and neck cancer. Front Oncol 2015;5:272.
Avanzo M, Stancanello J, El Naqa I. Beyond imaging: The promise of radiomics. Phys Med 2017;38:122-39.
Hinton GE, Osindero S, Teh YW. A fast learning algorithm for deep belief nets. Neural Comput 2006;18:1527-54.
Hinton GE, Salakhutdinov RR. Reducing the dimensionality of data with neural networks. Science 2006;313:504-7.
Kermany DS, Goldbaum M, Cai W, Valentim CC, Liang H, Baxter SL, et al.
Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell 2018;172:1122-31.