Content uploaded by Bin Zhang
Author content
All content in this area was uploaded by Bin Zhang on Mar 26, 2019
Content may be subject to copyright.
Radiomic Machine-Learning Classifiers for Prognostic Biomarkers of Advanced Nasopharyngeal Carcinoma
Abstract
We aimed to identify optimal machine-learning methods for radiomics-based prediction of local failure and distant failure in advanced nasopharyngeal carcinoma (NPC). We enrolled 110 patients with advanced NPC. A total of 970 radiomic features were extracted from MRI images for each patient. Six feature selection methods and nine classification methods were evaluated in terms of their performance. We applied the 10-fold cross-validation as the criterion for feature selection and classification. We repeated each combination for 50 times to obtain the mean area under the curve (AUC) and test error. We observed that the combination methods Random Forest (RF) + RF (AUC, 0.8464 ± 0.0069; test error, 0.3135 ± 0.0088) had the highest prognostic performance, followed by RF + Adaptive Boosting (AdaBoost) (AUC, 0.8204 ± 0.0095; test error, 0.3384 ± 0.0097), and Sure Independence Screening (SIS) + Linear Support Vector Machines (LSVM) (AUC, 0.7883 ± 0.0096; test error, 0.3985 ± 0.0100). Our radiomics study identified optimal machine-learning methods for the radiomics-based prediction of local failure and distant failure in advanced NPC, which could enhance the applications of radiomics in precision oncology and clinical practice.
... 12 Radiomic features are hand-crafted features, which are based on expert domain knowledge, including measures of intensity, shape, texture, and wavelet features. 13 Previous studies have shown that radiomic features might serve as potential biomarkers for prognosis in patients with NPC [14][15][16][17][18][19][20][21] ; however, they were associated with limitations of small sample size or lack of external validation that hindered their clinical utility. Besides, the period of survival predicted by all published prognostic models for NPC did not exceed 5 years. ...
... Besides, the period of survival predicted by all published prognostic models for NPC did not exceed 5 years. [14][15][16][17][18][19][20][21] Given that patient outcomes have been substantially improved because of the great advances in chemoradiotherapy strategies, these are urgently needed to construct prognostic models for predicting the survival of patients beyond 5 years. Compared with radiomics, deep learning technologies may better capture tumor heterogeneity, yet few studies to date have elucidated the prognostic value of feature representations automatically learned by deep learning from MRI images of NPC. ...
... 12,13 Previous radiomic studies have shown that MRI-derived biomarkers have the potential to provide information about treatment response and prognostication. [14][15][16][17][18][19][20][21] However, most studies were limited by the small sample size and the absence of external validation. Studies with a small sample size may prevent their findings from being extrapolated. ...
... This methodology extends the value of conventional MRI scans beyond anatomical visualization, enabling the quantification of tumor heterogeneity at a microscopic level that may not be visually apparent [5,6]. Machine learning further enhances this process by identifying complex relationships between radiomic features and clinical outcomes, facilitating the development of predictive models for NPC prognosis [7,8]. ...
... Table S2. After comprehensive analysis, 15 studies were included in our meta-analysis [21][22][23][24][25][26][27][28][29][30][31][32][33][34][35], with the exclusion rationale documented in Table S3 [8,10,34,. ...
These authors contributed equally to this work. Abstract: This meta-analysis investigates the prognostic value of MRI-based radiomics in nasopha-ryngeal carcinoma treatment outcomes, specifically focusing on overall survival (OS) variability. The study protocol was registered with INPLASY (INPLASY202420101). Initially, a systematic review identified 15 relevant studies involving 6243 patients through a comprehensive search across PubMed, Embase, and Web of Science, adhering to PRISMA guidelines. The methodological quality was assessed using the Quality in Prognosis Studies (QUIPS) tool and the Radiomics Quality Score (RQS), highlighting a low risk of bias in most domains. Our analysis revealed a significant average concordance index (c-index) of 72% across studies, indicating the potential of radiomics in clinical prognostication. However, moderate heterogeneity was observed, particularly in OS predictions. Subgroup analyses and meta-regression identified validation methods and radiomics software as significant heterogeneity moderators. Notably, the number of features in the prognosis model correlated positively with its performance. These findings suggest radiomics' promising role in enhancing cancer treatment strategies, though the observed heterogeneity and potential biases call for cautious interpretation and standardization in future research.
... One significant advantage of in silico studies is that researchers have no restrictions on the parameters or features they can use to find correlations. A radiomics study identified optimal ML methods for the radiomics-based prediction of local failure and distant failure in advanced nasopharyngeal carcinoma [60]. They tested different parameters and features in combination to divide patients into three groups: those who experienced local failure, those who experienced distant failure, and those who did not experience either local or distant failure during the follow-up period. ...
Simple Summary
This review examines the latest technologies using machine learning (ML) methods, including the use of convolutional neural networks, decision trees, and generative adversarial networks to solve problems concerning cancer recognition, planning optimal treatment strategies, as well as predicting the likelihood of patient survival. The authors also discuss the prospects of using machine learning technologies in medicine. In addition, the authors emphasize the need to address issues such as the anonymization of data received from patients.
Abstract
The role of machine learning (a part of artificial intelligence—AI) in the diagnosis and treatment of various types of oncology is steadily increasing. It is expected that the use of AI in oncology will speed up both diagnostic and treatment planning processes. This review describes recent applications of machine learning in oncology, including medical image analysis, treatment planning, patient survival prognosis, and the synthesis of drugs at the point of care. The fast and reliable analysis of medical images is of great importance in the case of fast-flowing forms of cancer. The introduction of ML for the analysis of constantly growing volumes of big data makes it possible to improve the quality of prescribed treatment and patient care. Thus, ML is expected to become an essential technology for medical specialists. The ML model has already improved prognostic prediction for patients compared to traditional staging algorithms. The direct synthesis of the necessary medical substances (small molecule mixtures) at the point of care could also seriously benefit from the application of ML. We further review the main trends in the use of artificial intelligence-based technologies in modern oncology. This review demonstrates the future prospects of using ML tools to make progress in cancer research, as well as in other areas of medicine. Despite growing interest in the use of modern computer technologies in medical practice, a number of unresolved ethical and legal problems remain. In this review, we also discuss the most relevant issues among them.
... Step by step, Machine learning applies in an ever increasing number of logical fields including, for instance, bioinformatics [5,6], natural chemistry [7] [8], medication [9] [10], meteorology [11] [12], financial sciences [13] [14], advanced mechanics [15] [16], hydroponics [17] [18], and food security [19] [20], and climatology [21]. ...
... Although the integration of feature screening methods and machine learning classifiers has reduced the dimensionality in big datasets, unsupervised clustering of imaging subtypes , c) pathologically confirmed LN status model and (e, f) nomogram in the training and external test sets, respectively. The nomogram showed significant differences for risk stratification in the total cohort and training set (log rank p < 0.05, respectively), but showed no statistical significance in the external test set (log rank p > 0.05, respectively) hinders accurate LN status prediction [32]. We combined the widely used mRMR and LASSO feature screening approaches with nine machine learning classifiers to determine the best radiomics machine learning signature to predict LN status. ...
Objective
To establish a model for predicting lymph node metastasis in bladder cancer (BCa) patients.
Methods
We retroactively enrolled 239 patients who underwent three-phase CT and resection for BCa in two centers (training set, n = 185; external test set, n = 54). We reviewed the clinical characteristics and CT features to identify significant predictors to construct a clinical model. We extracted the hand-crafted radiomics features and deep learning features of the lesions. We used the Minimum Redundancy Maximum Relevance algorithm and the least absolute shrinkage and selection operator logistic regression algorithm to screen features. We used nine classifiers to establish the radiomics machine learning signatures. To compensate for the uneven distribution of the data, we used the synthetic minority over-sampling technique to retrain each machine-learning classifier. We constructed the combined model using the top-performing radiomics signature and clinical model, and finally presented as a nomogram. We evaluated the combined model’s performance using the area under the receiver operating characteristic, accuracy, calibration curves, and decision curve analysis. We used the Kaplan–Meier survival curve to analyze the prognosis of BCa patients.
Results
The combined model incorporating radiomics signature and clinical model achieved an area under the receiver operating characteristic of 0.834 (95% CI: 0.659–1.000) for the external test set. The calibration curves and decision curve analysis demonstrated exceptional calibration and promising clinical use. The combined model showed good risk stratification performance for progression-free survival.
Conclusion
The proposed CT-based combined model is effective and reliable for predicting lymph node status of BCa patients preoperatively.
Critical relevance statement
Bladder cancer is a type of urogenital cancer that has a high morbidity and mortality rate. Lymph node metastasis is an independent risk factor for death in bladder cancer patients. This study aimed to investigate the performance of a deep learning radiomics model for preoperatively predicting lymph node metastasis in bladder cancer patients.
Key points
• Conventional imaging is not sufficiently accurate to determine lymph node status.
• Deep learning radiomics model accurately predicted bladder cancer lymph node metastasis.
• The proposed method showed satisfactory patient risk stratification for progression-free survival.
Graphical Abstract
Supplementary Information
The online version contains supplementary material available at 10.1186/s13244-023-01569-5.
Breakout is the most serious production accident in continuous casting and must be detected and predicted by stable and reliable methods. The sticking region, which forms on the local copper plate and expanded into a “V” shape, is the typical precursor of breakout. Therefore, computer vision technology was exploited to visualize the temperature change rate of the copper plate based on the temperature signals from thermocouples; then, the static and dynamic features of the abnormal sticking region were extracted. Meanwhile, logistic regression and Adaboost models were used to study and identify these features, resulting in the development of a mold breakout prediction model based on computer vision and machine learning. The test results demonstrate that the proposed model can effectively distinguish anomalous temperature patterns and considerably reduce false alarms without any missing reports. As a result, the proposed method could offer valuable insights into the realm of abnormality detection and prediction during continuous casting process.
Intelligent tutoring systems were developed to provide effective learning and educate learners without the human interventions. However, identifying learning disabilities is a challenge that requires an effective data collection process. As, the process of data collection affects the generated learner’s details and their learning outcomes in an intelligent tutoring system. Therefore, in the present work, we have compared the data collection process to record the response of the learning-disabled learners for the developed ITS. Also, the machine learning (ML) techniques were explored to extract and select the features related to learning disabilities (dyslexia, dysgraphia, and dyscalculia). This study includes comparison of some popular and effective feature selection and classification algorithms. The feature selection methods utilized in the present study are relief, XGBoost (XGB), elastic net (EN), least absolute shrinkage and selection operator (LASSO), and gradient boosting decision tree (GDBT). The classification algorithms that have been compared based on these feature selection techniques for learning disability prediction are K-nearest neighbors (KNN), logistic regression (LR), linear discriminant analysis (LDA), classification and regression trees (CART), Naïve Bayes (NB), and support vector machine (SVM). The result proves both data collection methods were equally effective and accuracy of CART with the feature selection method LASSO yields highest performance for feature selection and classification process.KeywordsClassificationFeature selectionLearning-difficulties
In this era of precision oncology, there has been an exponential growth in the armamentarium of genomically targeted therapies and immunotherapies. Evaluating early responses to precision therapy is essential for “go” versus “no go” decisions for these molecularly targeted drugs and agents that arm the immune system. Many different response assessment criteria exist for use in solid tumors and lymphomas. We reviewed the literature using the Medline/PubMed database for keywords “response assessment” and various known response assessment criteria published up to 2016. In this article we review the commonly used response assessment criteria. We present a decision tree to facilitate selection of appropriate criteria. We also suggest methods for standardization of various response assessment criteria. The relevant response assessment criteria were further studied for rational of development, key features, proposed use and acceptance by various entities. We also discuss early response evaluation and provide specific case studies of early response to targeted therapy. With high-throughput, advanced computing programs and digital data-mining it is now possible to acquire vast amount of high quality imaging data opening up a new field of “omics in radiology”—radiomics that complements genomics for personalized medicine. Radiomics is rapidly evolving and is still in the research arena. This cutting-edge technology is poised to move soon to the mainstream clinical arena. Novel agents with new mechanisms of action require advanced molecular imaging as imaging biomarkers. There is an urgent need for development of standardized early response assessment criteria for evaluation of response to precision therapy.
Medical image processing and analysis (also known as Radiomics) is a rapidly growing discipline that maps digital medical images into quantitative data, with the end goal of generating imaging biomarkers as decision support tools for clinical practice. The use of imaging data from routine clinical work-up has tremendous potential in improving cancer care by heightening understanding of tumor biology and aiding in the implementation of precision medicine. As a noninvasive method of assessing the tumor and its microenvironment in their entirety, radiomics allows the evaluation and monitoring of tumor characteristics such as temporal and spatial heterogeneity. One can observe a rapid increase in the number of computational medical imaging publications-milestones that have highlighted the utility of imaging biomarkers in oncology. Nevertheless, the use of radiomics as clinical biomarkers still necessitates amelioration and standardization in order to achieve routine clinical adoption. This Review addresses the critical issues to ensure the proper development of radiomics as a biomarker and facilitate its implementation in clinical practice. © The Author 2017. Published by Oxford University Press on behalf of the European Society for Medical Oncology. All rights reserved.
Objectives
Radiomics utilizes quantitative image features (QIFs) to characterize tumor phenotype. In practice, radiological images are obtained from different vendors’ equipment using various imaging acquisition settings. Our objective was to assess the inter-setting agreement of QIFs computed from CT images by varying two parameters, slice thickness and reconstruction algorithm.
Materials and Methods
CT images from an IRB-approved/HIPAA-compliant study assessing thirty-two lung cancer patients were included for the analysis. Each scan’s raw data were reconstructed into six imaging series using combinations of two reconstruction algorithms (Lung[L] and Standard[S]) and three slice thicknesses (1.25mm, 2.5mm and 5mm), i.e., 1.25L, 1.25S, 2.5L, 2.5S, 5L and 5S. For each imaging-setting, 89 well-defined QIFs were computed for each of the 32 tumors (one tumor per patient). The six settings led to 15 inter-setting comparisons (combinatorial pairs). To reduce QIF redundancy, hierarchical clustering was done. Concordance correlation coefficients (CCCs) were used to assess inter-setting agreement of the non-redundant feature groups. The CCC of each group was assessed by averaging CCCs of QIFs in the group.
Results
Twenty-three non-redundant feature groups were created. Across all feature groups, the best inter-setting agreements (CCCs>0.8) were 1.25S vs 2.5S, 1.25L vs 2.5L, and 2.5S vs 5S; the worst (CCCs<0.51) belonged to 1.25L vs 5S and 2.5L vs 5S. Eight of the feature groups related to size, shape, and coarse texture had an average CCC>0.8 across all imaging settings.
Conclusions
Varying degrees of inter-setting disagreements of QIFs exist when features are computed from CT images reconstructed using different algorithms and slice thicknesses. Our findings highlight the importance of harmonizing imaging acquisition for obtaining consistent QIFs to study tumor imaging phonotype.
PurposeTo evaluate the associations between clinical outcomes and radiomics-derived inter-site spatial heterogeneity metrics across multiple metastatic lesions on CT in patients with high-grade serous ovarian cancer (HGSOC). MethodsIRB-approved retrospective study of 38 HGSOC patients. All sites of suspected HGSOC involvement on preoperative CT were manually segmented. Gray-level correlation matrix-based textures were computed from each tumour site, and grouped into five clusters using a Gaussian Mixture Model. Pairwise inter-site similarities were computed, generating an inter-site similarity matrix (ISM). Inter-site texture heterogeneity metrics were computed from the ISM and compared to clinical outcomes. ResultsOf the 12 inter-site texture heterogeneity metrics evaluated, those capturing the differences in texture similarities across sites were associated with shorter overall survival (inter-site similarity entropy, similarity level cluster shade, and inter-site similarity level cluster prominence; p ≤ 0.05) and incomplete surgical resection (similarity level cluster shade, inter-site similarity level cluster prominence and inter-site cluster variance; p ≤ 0.05). Neither the total number of disease sites per patient nor the overall tumour volume per patient was associated with overall survival. Amplification of 19q12 involving cyclin E1 gene (CCNE1) predominantly occurred in patients with more heterogeneous inter-site textures. Conclusion
Quantitative metrics non-invasively capturing spatial inter-site heterogeneity may predict outcomes in patients with HGSOC. Key Points• Calculating inter-site texture-based heterogeneity metrics was feasible• Metrics capturing texture similarities across HGSOC sites were associated with overall survival• Heterogeneity metrics were also associated with incomplete surgical resection of HGSOC.
Background:
To stratify risks of pancreatic adenocarcinoma (PA) patients using pre- and post-radiotherapy (RT) PET/CT images, and to assess the prognostic value of texture variations in predicting therapy response of patients.
Methods:
Twenty-six PA patients treated with RT from 2011-2013 with pre- and post-treatment 18F-FDG-PET/CT scans were identified. Tumor locoregional texture was calculated using 3D kernel-based approach, and texture variations were identified by fitting discrepancies of texture maps of pre- and post-treatment images. A total of 48 texture and clinical variables were identified and evaluated for association with overall survival (OS). The prognostic heterogeneity features were selected using lasso/elastic net regression, and further were evaluated by multivariate Cox analysis.
Results:
Median age was 69 y (range, 46-86 y). The texture map and temporal variations between pre- and post-treatment were well characterized by histograms and statistical fitting. The lasso analysis identified seven predictors (age, node stage, post-RT SUVmax, variations of homogeneity, variance, sum mean, and cluster tendency). The multivariate Cox analysis identified five significant variables: age, node stage, variations of homogeneity, variance, and cluster tendency (with P=0.020, 0.040, 0.065, 0.078, and 0.081, respectively). The patients were stratified into two groups based on the risk score of multivariate analysis with log-rank P=0.001: a low risk group (n=11) with a longer mean OS (29.3 months) and higher texture variation (>30%), and a high risk group (n=15) with a shorter mean OS (17.7 months) and lower texture variation (<15%).
Conclusions:
Locoregional metabolic texture response provides a feasible approach for evaluating and predicting clinical outcomes following treatment of PA with RT. The proposed method can be used to stratify patient risk and help select appropriate treatment strategies for individual patients toward implementing response-driven adaptive RT.
Purpose:
To investigate the biological meaning of apparent diffusion coefficient (ADC) values in tumors following radiotherapy.
Materials and methods:
Five mice bearing TRAMP-C1 tumor were half-irradiated with a dose of 15 Gy. Diffusion-weighted images, using multiple b-values from 0 to 3000 s/mm2 , were acquired at 7T on day 6. ADC values calculated by a two-point estimate and monoexponential fitting of signal decay were compared between the irradiated and nonirradiated regions of the tumor. Pixelwise ADC maps were correlated with histological metrics including nuclear counts, nuclear sizes, nuclear spaces, cytoplasmic spaces, and extracellular spaces.
Results:
As compared with the nonirradiated region, the irradiated region exhibited significant increases in ADC, extracellular space, and nuclear size, and a significant decrease in nuclear counts (P < 0.001 for all). Optimal ADC to differentiate the irradiated from nonirradiated regions was achieved at a b-value of 800 s/mm2 by the two-point method and monoexponential curve fitting. ADC positively correlated with extracellular spaces (r = 0.74) and nuclear sizes (r = 0.72), and negatively correlated with nuclear counts (r = -0.82, P < 0.001 for all).
Conclusion:
As a radiomic biomarker, ADC maps correlating with histological metrics pixelwise could be a means of evaluating tumor heterogeneity and responses to radiotherapy.
Level of evidence:
1 Technical Efficacy: Stage 2 J. MAGN. RESON. IMAGING 2017;46:483-489.
Purpose:
Many radiomics features were originally developed for non-medical imaging applications and therefore original assumptions may need to be reexamined. In this study, we investigated the impact of slice thickness and pixel spacing (or pixel size) on radiomics features extracted from Computed Tomography (CT) phantom images acquired with different scanners as well as different acquisition and reconstruction parameters. The dependence of CT texture features on gray-level discretization was also evaluated.
Methods and materials:
A texture phantom composed of 10 different cartridges of different materials was scanned on eight different CT scanners from three different manufacturers. The images were reconstructed for various slice thicknesses. For each slice thickness, the reconstruction Field Of View (FOV) was varied to render pixel sizes ranging from 0.39 to 0.98 mm. A fixed spherical region of interest (ROI) was contoured on the images of the shredded rubber cartridge and the 3D printed, 20% fill, acrylonitrile butadiene styrene plastic cartridge (ABS20) for all phantom imaging sets. Radiomic features were extracted from the ROIs using an in-house program. Features categories were: shape (10), intensity (16), GLCM (24), GLZSM (11), GLRLM (11), and NGTDM (5), fractal dimensions (8) and first-order wavelets (128), for a total of 213 features. Voxel-size resampling was performed to investigate the usefulness of extracting features using a suitably chosen voxel size. Acquired phantom image sets were resampled to a voxel size of 1 × 1 × 2 mm(3) using linear interpolation. Image features were therefore extracted from resampled and original datasets and the absolute value of the percent coefficient of variation (%COV) for each feature was calculated. Based on the %COV values, features were classified in 3 groups: (1) features with large variations before and after resampling (%COV >50); (2) features with diminished variation (%COV <30) after resampling; and (3) features that had originally moderate variation (%COV <50%) and were negligibly affected by resampling. Group 2 features were further studied by modifying feature definitions to include voxel size. Original and voxel-size normalized features were used for interscanner comparisons. A subsequent analysis investigated feature dependency on gray-level discretization by extracting 51 texture features from ROIs from each of the 10 different phantom cartridges using 16, 32, 64, 128, and 256 gray levels.
Results:
Out of the 213 features extracted, 150 were reproducible across voxel sizes, 42 improved significantly (%COV <30, Group 2) after resampling, and 21 had large variations before and after resampling (Group 1). Ten features improved significantly after definition modification effectively removed their voxel-size dependency. Interscanner comparison indicated that feature variability among scanners nearly vanished for 8 of these 10 features. Furthermore, 17 out of 51 texture features were found to be dependent on the number of gray levels. These features were redefined to include the number of gray levels which greatly reduced this dependency.
Conclusion:
Voxel-size resampling is an appropriate pre-processing step for image datasets acquired with variable voxel sizes to obtain more reproducible CT features. We found that some of the radiomics features were voxel size and gray-level discretization-dependent. The introduction of normalizing factors in their definitions greatly reduced or removed these dependencies.
Radiomics is a new area of research in the field of imaging with tremendous potential to unravel the hidden information in digital images. The scope of radiology has grown exponentially over the last two decades; since the advent of radiomics, many quantitative imaging features can now be extracted from medical images through high-throughput computing, and these can be converted into mineable data that can help in linking imaging phenotypes with clinical data, genomics, proteomics, and other "omics" information. In cancer, radiomic imaging analysis aims at extracting imaging features embedded in the imaging data, which can act as a guide in the disease or cancer diagnosis, staging and planning interventions for treating patients, monitor patients on therapy, predict treatment response, and determine patient outcomes.
Objective
The status of isocitrate dehydrogenase 1 (IDH1) is highly correlated with the development, treatment and prognosis of glioma. We explored a noninvasive method to reveal IDH1 status by using a quantitative radiomics approach for grade II glioma. MethodsA primary cohort consisting of 110 patients pathologically diagnosed with grade II glioma was retrospectively studied. The radiomics method developed in this paper includes image segmentation, high-throughput feature extraction, radiomics sequencing, feature selection and classification. Using the leave-one-out cross-validation (LOOCV) method, the classification result was compared with the real IDH1 situation from Sanger sequencing. Another independent validation cohort containing 30 patients was utilised to further test the method. ResultsA total of 671 high-throughput features were extracted and quantized. 110 features were selected by improved genetic algorithm. In LOOCV, the noninvasive IDH1 status estimation based on the proposed approach presented an estimation accuracy of 0.80, sensitivity of 0.83 and specificity of 0.74. Area under the receiver operating characteristic curve reached 0.86. Further validation on the independent cohort of 30 patients produced similar results. Conclusions
Radiomics is a potentially useful approach for estimating IDH1 mutation status noninvasively using conventional T2-FLAIR MRI images. The estimation accuracy could potentially be improved by using multiple imaging modalities. Key Points• Noninvasive IDH1 status estimation can be obtained with a radiomics approach.• Automatic and quantitative processes were established for noninvasive biomarker estimation.• High-throughput MRI features are highly correlated to IDH1 states.• Area under the ROC curve of the proposed estimation method reached 0.86.