Purpose
Machine learning classification algorithms (classifiers) for prediction of treatment response are becoming more popular in radiotherapy literature. General Machine learning literature ...provides evidence in favor of some classifier families (random forest, support vector machine, gradient boosting) in terms of classification performance. The purpose of this study is to compare such classifiers specifically for (chemo)radiotherapy datasets and to estimate their average discriminative performance for radiation treatment outcome prediction.
Methods
We collected 12 datasets (3496 patients) from prior studies on post‐(chemo)radiotherapy toxicity, survival, or tumor control with clinical, dosimetric, or blood biomarker features from multiple institutions and for different tumor sites, that is, (non‐)small‐cell lung cancer, head and neck cancer, and meningioma. Six common classification algorithms with built‐in feature selection (decision tree, random forest, neural network, support vector machine, elastic net logistic regression, LogitBoost) were applied on each dataset using the popular open‐source R package caret. The R code and documentation for the analysis are available online (https://github.com/timodeist/classifier_selection_code). All classifiers were run on each dataset in a 100‐repeated nested fivefold cross‐validation with hyperparameter tuning. Performance metrics (AUC, calibration slope and intercept, accuracy, Cohen's kappa, and Brier score) were computed. We ranked classifiers by AUC to determine which classifier is likely to also perform well in future studies. We simulated the benefit for potential investigators to select a certain classifier for a new dataset based on our study (pre‐selection based on other datasets) or estimating the best classifier for a dataset (set‐specific selection based on information from the new dataset) compared with uninformed classifier selection (random selection).
Results
Random forest (best in 6/12 datasets) and elastic net logistic regression (best in 4/12 datasets) showed the overall best discrimination, but there was no single best classifier across datasets. Both classifiers had a median AUC rank of 2. Preselection and set‐specific selection yielded a significant average AUC improvement of 0.02 and 0.02 over random selection with an average AUC rank improvement of 0.42 and 0.66, respectively.
Conclusion
Random forest and elastic net logistic regression yield higher discriminative performance in (chemo)radiotherapy outcome and toxicity prediction than other studied classifiers. Thus, one of these two classifiers should be the first choice for investigators when building classification models or to benchmark one's own modeling results against. Our results also show that an informed preselection of classifiers based on existing datasets can improve discrimination over random selection.
Artificial intelligence (AI) is a general term that implies the use of a computer to model intelligent behavior with minimal human intervention. AI, particularly deep learning, has recently made ...substantial strides in perception tasks allowing machines to better represent and interpret complex data. Deep learning is a subset of AI represented by the combination of artificial neuron layers. In the last years, deep learning has gained great momentum. In the field of orthopaedics and traumatology, some studies have been done using deep learning to detect fractures in radiographs. Deep learning studies to detect and classify fractures on computed tomography (CT) scans are even more limited. In this narrative review, we provide a brief overview of deep learning technology: we (1) describe the ways in which deep learning until now has been applied to fracture detection on radiographs and CT examinations; (2) discuss what value deep learning offers to this field; and finally (3) comment on future directions of this technology.
Celotno besedilo
Dostopno za:
DOBA, FSPLJ, IZUM, KILJ, NUK, OILJ, PILJ, PNG, SAZU, UILJ, UKNU, UL, UM, UPUK, VSZLJ
FDG-PET-derived textural features describing intra-tumor heterogeneity are increasingly investigated as imaging biomarkers. As part of the process of quantifying heterogeneity, image intensities ...(SUVs) are typically resampled into a reduced number of discrete bins. We focused on the implications of the manner in which this discretization is implemented. Two methods were evaluated: (1) R(D), dividing the SUV range into D equally spaced bins, where the intensity resolution (i.e. bin size) varies per image; and (2) R(B), maintaining a constant intensity resolution B. Clinical feasibility was assessed on 35 lung cancer patients, imaged before and in the second week of radiotherapy. Forty-four textural features were determined for different D and B for both imaging time points. Feature values depended on the intensity resolution and out of both assessed methods, R(B) was shown to allow for a meaningful inter- and intra-patient comparison of feature values. Overall, patients ranked differently according to feature values–which was used as a surrogate for textural feature interpretation–between both discretization methods. Our study shows that the manner of SUV discretization has a crucial effect on the resulting textural features and the interpretation thereof, emphasizing the importance of standardized methodology in tumor texture analysis.
Medical imaging can visualize characteristics of human cancer noninvasively. Radiomics is an emerging field that translates these medical images into quantitative data to enable phenotypic profiling ...of tumors. While radiomics has been associated with several clinical endpoints, the complex relationships of radiomics, clinical factors, and tumor biology are largely unknown. To this end, we analyzed two independent cohorts of respectively 262 North American and 89 European patients with lung cancer, and consistently identified previously undescribed associations between radiomic imaging features, molecular pathways, and clinical factors. In particular, we found a relationship between imaging features, immune response, inflammation, and survival, which was further validated by immunohistochemical staining. Moreover, a number of imaging features showed predictive value for specific pathways; for example, intra-tumor heterogeneity features predicted activity of RNA polymerase transcription (AUC = 0.62, p=0.03) and intensity dispersion was predictive of the autodegration pathway of a ubiquitin ligase (AUC = 0.69, p
10
). Finally, we observed that prognostic biomarkers performed highest when combining radiomic, genetic, and clinical information (CI = 0.73, p<10
) indicating complementary value of these data. In conclusion, we demonstrate that radiomic approaches permit noninvasive assessment of both molecular and clinical characteristics of tumors, and therefore have the potential to advance clinical decision-making by systematically analyzing standard-of-care medical images.
Historically, medical imaging has been a qualitative or semi-quantitative modality. It is difficult to quantify what can be seen in an image, and to turn it into valuable predictive outcomes. As a ...result of advances in both computational hardware and machine learning algorithms, computers are making great strides in obtaining quantitative information from imaging and correlating it with outcomes. Radiomics, in its two forms "handcrafted and deep," is an emerging field that translates medical images into quantitative data to yield biological information and enable radiologic phenotypic profiling for diagnosis, theragnosis, decision support, and monitoring. Handcrafted radiomics is a multistage process in which features based on shape, pixel intensities, and texture are extracted from radiographs. Within this review, we describe the steps: starting with quantitative imaging data, how it can be extracted, how to correlate it with clinical and biological outcomes, resulting in models that can be used to make predictions, such as survival, or for detection and classification used in diagnostics. The application of deep learning, the second arm of radiomics, and its place in the radiomics workflow is discussed, along with its advantages and disadvantages. To better illustrate the technologies being used, we provide real-world clinical applications of radiomics in oncology, showcasing research on the applications of radiomics, as well as covering its limitations and its future direction.
This study investigated an association of post-radiochemotherapy (RCT) PET radiomics with local tumor control in head and neck squamous cell carcinoma (HNSCC) and evaluated the models against two ...radiomics software implementations.
649 features, available in two radiomics implementations and based on the same definitions, were extracted from HNSCC primary tumor region in 18F-FDG PET scans 3 months post definitive RCT (training cohort n = 128, validation cohort n = 50) and compared using the intraclass correlation coefficient (ICC). Local recurrence models were trained, separately for both implementations, using principal component analysis (PCA) and the least absolute shrinkage and selection operator. The reproducibility of the concordance indexes (CI) in univariable Cox regression for features preselected in PCA and the final multivariable models was investigated using respective features from the other implementation.
Only 80 PET radiomic features yielded ICC > 0.8 in the comparison between the implementations. The change of implementation caused high variability of CI in the univariable analysis. However, both final multivariable models performed equally well in the training and validation cohorts (CI > 0.7) independent of radiomics implementation.
The two post-RCT PET radiomic models, based on two different software implementations, were prognostic for local tumor control in HNSCC. However, 88% of the features was not reproducible between the implementations.
Two CT features were developed to quantitatively describe lung adenocarcinomas by scoring tumor shape complexity (feature 1: convexity) and intratumor density variation (feature 2: entropy ratio) in ...routinely obtained diagnostic CT scans. The developed quantitative features were analyzed in two independent cohorts (cohort 1: n = 61; cohort 2: n = 47) of patients diagnosed with primary lung adenocarcinoma, retrospectively curated to include imaging and clinical data. Preoperative chest CTs were segmented semi-automatically. Segmented tumor regions were further subdivided into core and boundary sub-regions, to quantify intensity variations across the tumor. Reproducibility of the features was evaluated in an independent test-retest dataset of 32 patients. The proposed metrics showed high degree of reproducibility in a repeated experiment (concordance, CCC≥0.897; dynamic range, DR≥0.92). Association with overall survival was evaluated by Cox proportional hazard regression, Kaplan-Meier survival curves, and the log-rank test. Both features were associated with overall survival (convexity: p = 0.008; entropy ratio: p = 0.04) in Cohort 1 but not in Cohort 2 (convexity: p = 0.7; entropy ratio: p = 0.8). In both cohorts, these features were found to be descriptive and demonstrated the link between imaging characteristics and patient survival in lung adenocarcinoma.
Celotno besedilo
Dostopno za:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Tumor hypoxia is a common microenvironmental factor that adversely influences tumor phenotype and treatment response. Cellular adaptation to hypoxia occurs through multiple mechanisms, including ...activation of the unfolded protein response (UPR). Recent reports have indicated that hypoxia activates a lysosomal degradation pathway known as autophagy, and here we show that the UPR enhances the capacity of hypoxic tumor cells to carry out autophagy, and that this promotes their survival. In several human cancer cell lines, hypoxia increased transcription of the essential autophagy genes microtubule-associated protein 1 light chain 3beta (MAP1LC3B) and autophagy-related gene 5 (ATG5) through the transcription factors ATF4 and CHOP, respectively, which are regulated by PKR-like ER kinase (PERK, also known as EIF2AK3). MAP1LC3B and ATG5 are not required for initiation of autophagy but mediate phagophore expansion and autophagosome formation. We observed that transcriptional induction of MAP1LC3B replenished MAP1LC3B protein that was turned over during extensive hypoxia-induced autophagy. Correspondingly, cells deficient in PERK signaling failed to transcriptionally induce MAP1LC3B and became rapidly depleted of MAP1LC3B protein during hypoxia. Consistent with these data, autophagy and MAP1LC3B induction occurred preferentially in hypoxic regions of human tumor xenografts. Furthermore, pharmacological inhibition of autophagy sensitized human tumor cells to hypoxia, reduced the fraction of viable hypoxic tumor cells, and sensitized xenografted human tumors to irradiation. Our data therefore demonstrate that the UPR is an important mediator of the hypoxic tumor microenvironment and that it contributes to resistance to treatment through its ability to facilitate autophagy.
Abstract Electronic portal imaging devices (EPIDs) have been the preferred tools for verification of patient positioning for radiotherapy in recent decades. Since EPID images contain dose ...information, many groups have investigated their use for radiotherapy dose measurement. With the introduction of the amorphous-silicon EPIDs, the interest in EPID dosimetry has been accelerated because of the favourable characteristics such as fast image acquisition, high resolution, digital format, and potential for in vivo measurements and 3D dose verification. As a result, the number of publications dealing with EPID dosimetry has increased considerably over the past ∼15 years. The purpose of this paper was to review the information provided in these publications. Information available in the literature included dosimetric characteristics and calibration procedures of various types of EPIDs, strategies to use EPIDs for dose verification, clinical approaches to EPID dosimetry, ranging from point dose to full 3D dose distribution verification, and current clinical experience. Quality control of a linear accelerator, pre-treatment dose verification and in vivo dosimetry using EPIDs are now routinely used in a growing number of clinics. The use of EPIDs for dosimetry purposes has matured and is now a reliable and accurate dose verification method that can be used in a large number of situations. Methods to integrate 3D in vivo dosimetry and image-guided radiotherapy (IGRT) procedures, such as the use of kV or MV cone-beam CT, are under development. It has been shown that EPID dosimetry can play an integral role in the total chain of verification procedures that are implemented in a radiotherapy department. It provides a safety net for simple to advanced treatments, as well as a full account of the dose delivered. Despite these favourable characteristics and the vast range of publications on the subject, there is still a lack of commercially available solutions for EPID dosimetry. As strategies evolve and commercial products become available, EPID dosimetry has the potential to become an accurate and efficient means of large-scale patient-specific IMRT dose verification for any radiotherapy department.
Radiomic analyses of CT images provide prognostic information that can potentially be used for personalized treatment. However, heterogeneity of acquisition- and reconstruction protocols influences ...robustness of radiomic analyses. The aim of this study was to investigate the influence of different CT-scanners, slice thicknesses, exposures and gray-level discretization on radiomic feature values and their stability.
A texture phantom with ten different inserts was scanned on nine different CT-scanners with varying tube currents. Scans were reconstructed with 1.5 mm or 3 mm slice thickness. Image pre-processing comprised gray-level discretization in ten different bin widths ranging from 5 to 50 HU and different resampling methods (i.e., linear, cubic and nearest neighbor interpolation to 1 × 1 × 3 mm
voxels) were investigated. Subsequently, 114 textural radiomic features were extracted from a 2.1 cm
sphere in the center of each insert. The influence of slice thickness, exposure and bin width on feature values was investigated. Feature stability was assessed by calculating the concordance correlation coefficient (CCC) in a test-retest setting and for different combinations of scanners, tube currents and slice thicknesses.
Bin width influenced feature values, but this only had a marginal effect on the total number of stable features (CCC > 0.85) when comparing different scanners, slice thicknesses or exposures. Most radiomic features were affected by slice thickness, but this effect could be reduced by resampling the CT-images before feature extraction. Statistics feature 'energy' was the most dependent on slice thickness. No clear correlation between feature values and exposures was observed.
CT-scanner, slice thickness and bin width affected radiomic feature values, whereas no effect of exposure was observed. Optimization of gray-level discretization to potentially improve prognostic value can be performed without compromising feature stability. Resampling images prior to feature extraction decreases the variability of radiomic features.
Celotno besedilo
Dostopno za:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, UILJ, UKNU, UL, UM, UPUK