Machine Learning methods for Quantitative Radiomic Biomarkers.
Parmar C, Grossmann P, Bussink J, Lambin P, Aerts HJWL. · 2015
Read Machine Learning methods for Quantitative Radiomic Biomarkers. by Parmar C, Grossmann P, Bussink J, Lambin P, Aerts HJWL. (2015) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4538374.txt.
Introduction
Chintan Parmar and four co-authors run a bake-off of machine learning methods for tumor scan data, in Scientific Reports in 2015. Radiomics extracts hundreds of numbers from a tumor's scan to describe its shape and texture. Whether those numbers can forecast survival depends on the learning method picked, a choice often made by habit.
From scans of 464 lung cancer patients, the team extracted 440 features, then tested 14 ways to select features and 12 ways to classify patients. Separate groups of 310 and 154 patients served for training and validation. A Wilcoxon-based selector and a random forest classifier came out most accurate and most steady under data wobbles. The classifier choice drove more of the outcome spread than any other step.
Catalog record
| CREATOR | Parmar C, Grossmann P, Bussink J, Lambin P, Aerts HJWL. |
|---|---|
| YEAR | 2015 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Machine Learning methods for Quantitative Radiomic Biomarkers By Chintan Parmar, Patrick Grossmann, Johan Bussink, Philippe Lambin, Hugo J W L Aerts Published in: Scientific Reports DOI: 10.1038/srep13087 ABSTRACT Abstract Radiomics extracts and mines large number of medical imaging features quantifying tumor phenotypic characteristics. Highly accurate and reliable machine-learning approaches can drive the success of radiomic applications in clinical care. In this radiomic study, fourteen feature selection methods and twelve classification methods were examined in terms of their performance and stability for predicting overall survival. A total of 440 radiomic features were extracted from pre-treatment computed tomography (CT) images of 464 lung cancer patients. To ensure the unbiased evaluation of different machine-learning methods, publicly available implementations along with reported parameter configurations were used. Furthermore, we used two independent radiomic cohorts for training (n = 310 patients) and validation (n = 154 patients). We identified that Wilcoxon test based feature selection method WLCX (stability = 0.84 ± 0.05, AUC = 0.65 ± 0.02) and a classification method random forest RF (RSD = 3.52%, AUC = 0.66 ± 0.03) had highest prognostic performance with high stability against data perturbation. Our variability analysis indicated that the choice of classification method is the most dominant source of performance variation (34.21% of total variance). Identification of optimal machine-learning methods for radiomic applications is a crucial step towards stable and clinically relevant radiomic biomarkers, providing a non-invasive way of quantifying and monitoring tumor-phenotypic characteristics in clinical practice. ‘Precision oncology’ refers to the customization of cancer care, where practices and/or therapies are being tailored to individual patients. Such customization process can maximize the success of preventive and therapeutic interventions with minimum side effects. Most of the precision oncology related research has centered on the molecular characterization of tumors using genomics based approaches, which require tissue extraction by tumor biopsies. Although several genomics based approaches have successfully been applied in clinical oncology1, there are inherent limitations to biopsy based assays. Tumors are spatially and temporally heterogeneous, and repeated tumor biopsies, which increase the risk for a patient, are often required to capture the molecular heterogeneity of tumors. These ethical and clinical challenges related to biopsy-based assays, can be addressed by medical imaging, which is a routine practice for cancer diagnosis and staging in clinical oncology. Unlike biopsies, medical imaging is non-invasive and can provide information regarding the entire tumor phenotype, including the intra-tumor heterogeneity. Furthermore, recent advances in high-resolution image acquisition machines and computational hardware allow the detailed and efficient quantification of tumor phenotypic characteristics. Therefore, medical imaging provides unprecedented opportunities for precision oncology. “Radiomics”, an emerging and promising field, hypothesizes that medical imaging provides crucial information regarding tumor physiology, which could be exploited to enhance cancer diagnostics2. It provides a comprehensive quantification of tumor phenotypes by extracting and mining large number of quantitative imaging features3. Several studies have investigated various radiomic features in terms of their prognostic or predictive abilities and reliability across different clinical settings4,5,6,7,8,9,10. Different studies have shown the discriminating capabilities of radiomic features for the stratification of tumor histology6, tumor grades or stages11, and clinical outcomes8,12,13. Moreover, some studies have reported the association between radiomic features and the underlying gene expression patterns8,14,15. “Machine-learning” can be broadly defined as computational methods/models using experience (data) to improve performance or make accurate predictions16. These programmable computational methods are capable of “learning” from data and hence can automate and improve the prediction process. Predictive and prognostic models with high accuracy, reliability, and efficiency are vital factors driving the success of radiomics. Therefore, it is essential to compare different machine-learning models for radiomics based clinical biomarkers. Like any high-throughput data-mining field, radiomics also underlies the curse of dimensionality17, which should be addressed by appropriate feature selection strategies. Moreover, feature selection also helps in reducing overfitting of models (increasing the generalizability). Thus, in order to reduce the dimensionality of radiomic feature space and enhance the performance of radiomics based predictive models, different feature selection methods18 should be thoroughly investigated. However, as radiomics is an emerging research field, most of the published studies have only assessed the predictive capabilities of radiomic features without putting much emphasis on the comparison of different feature selection and predictive modeling methods. Only few recent studies have investigated the effect of different feature selection and machine learning classification methods on radiomics based clinical predictions19,20, but with limited sample sizes. Furthermore, these studies lacked independent validation of the results, which may restrict the generalizability of their conclusions. In this study, we investigated a large panel of machine-learning approaches for radiomics based survival prediction. We evaluated 14 feature selection methods and 12 classification methods in terms of their predictive performance and stability against data perturbation. These methods were chosen because of their popularity in literature. Furthermore, publicly available implementations along with reported parameter configurations were used in the analysis, which ensured an unbiased evaluation of these methods. Two independent lung cancer cohorts were used for training and validation, with in total image and clinical outcome data of 464 patients. Feature selection and predictive modeling are considered as the important building blocks for high throughput data driven radiomics. Therefore, our investigation could help in the identification of optimal machine-learning approaches for radiomics based predictive studies, which could enhance the applications of non-invasive and cost-effective radiomics in clinical oncology. METHODS Radiomic Features A total of 440 radiomic features were used in the analysis. These radiomic features quantified tumor phenotypic characteristics on CT images and are divided into four feature groups: I) tumor intensity, II) shape, III) texture and IV) wavelet features. Tumor intensity based features estimated the first order statistics of the intensity histogram, whereas shape features described the 3D geometric properties of the tumor. Textural features, derived from the gray level co-occurrence (GLCM)21 and run length matrices (GLRLM)22, quantified the intra-tumor heterogeneity. These textural features were computed by averaging their values over all thirteen directions. Wavelet features are the transformed domain representations of the intensity and textural features. These features were computed on different wavelet decompositions of the original image using a coiflet wavelet transformation. Matlab R2012b (The Mathworks, Natick, MA) was used for the image analysis. Radiomic features were automatically extracted by our in-house developed radiomics image analysis software, which uses an adapted version of CERR (Computational Environment for Radiotherapy Research)23 and Matlab for the preprocessing of medical images. Mathematical definitions of all radiomic features, as well as the extraction methods, were previously described8. Datasets In this study, we employed two NSCLC cohorts from the two different institutes of Netherlands: (1) Lung1:422 NSCLC patients treated at MAASTRO Clinic in Maastricht. (2) Lung2:225 NSCLC patients treated at Radboud University Medical Center in Nijmegen. CT-scans, manual delineations and clinical data were available for all included patients. More details on the included datasets are described in Supplementary-A. We dichotomized the censored continuous survival data using a cutoff time of 2 years. The patients who lived beyond the cutoff time were labeled as 1, whereas the deceased ones were labeled as 0. The objective of the study was to stratify patients into these two labeled survival classes. Two-years is considered as a relevant survival time for NSCLC patients and several other studies have designed their prediction models using a survival cutoff of 2