Artificial Intelligence-Guided Prediction of Dental Doses Before Planning of Radiation Therapy for Oropharyngeal Cancer: Technical Development and Initial Feasibility of Implementation.

Purpose: The aim was to develop a novel artificial intelligence (AI)-guided clinical decision support system, to predict radiation doses to subsites of the mandible using diagnostic computed tomography scans acquired before any planning of head and neck radiation therapy (RT). Methods and Materials: A dose classifier was trained using RT plans from 86 patients with oropharyngeal cancer; the test set consisted of an additional 20 plans. The classifier was trained to predict whether mandible subsites would receive a mean dose >50 Gy. The AI predictions were prospectively evaluated and compared with those of a specialist head and neck radiation oncologist for 9 patients. Positive predictive value (PPV), negative predictive value (NPV), Pearson correlation coefficient, and Lin concordance correlation coefficient were calculated to compare the AI predictions to those of the physician. Results: In the test data set, the AI predictions had a PPV of 0.95 and NPV of 0.88. For 9 patients evaluated prospectively, there was a strong correlation between the predictions of the AI algorithm and physician (P = .72, P < .001). Comparing the AI algorithm versus the physician, the PPVs were 0.82 versus 0.25, and the NPVs were 0.94 versus 1.0, respectively. Concordance between physician estimates and final planned doses was 0.62; this was 0.71 between AI-based estimates and final planned doses. Conclusion: AI-guided decision support increased precision and accuracy of pre-RT dental dose estimates.

Stereotactic Radiosurgery for Vestibular Schwannomas: Tumor Control Probability Analyses and Recommended Reporting Standards.

PURPOSE: We sought to investigate the tumor control probability (TCP) of vestibular schwannomas after single-fraction stereotactic radiosurgery (SRS) or hypofractionated SRS over 2 to 5 fractions (fSRS). METHODS AND MATERIALS: Studies (PubMed indexed from 1993-2017) were eligible for data extraction if they contained dosimetric details of SRS/fSRS correlated with local tumor control. The rate of tumor control at 5 years (or at 3 years if 5-year data were not available) were collated. Poisson modeling estimated the TCP per equivalent dose in 2 Gy per fraction (EQD2) and in 1, 3, and 5 fractions. RESULTS: Data were extracted from 35 publications containing a total of 5162 patients. TCP modeling was limited by the absence of analyzable data of <11 Gy in a single-fraction, variability in definition of "tumor control," and by lack of significant increase in TCP for doses >12 Gy. Using linear-quadratic-based dose conversion, the 3- to 5-year TCP was estimated at 95% at an EQD2 of 25 Gy, corresponding to 1-, 3-, and 5-fraction doses of 13.8 Gy, 19.2 Gy, and 21.5 Gy, respectively. Single-fraction doses of 10 Gy, 11 Gy, 12 Gy, and 13 Gy predicted a TCP of 85.0%, 88.4%, 91.2%, and 93.5%, respectively. For fSRS, 18 Gy in 3 fractions (EQD2 of 23.0 Gy) and 25 Gy in 5 fractions (EQD2 of 30.2 Gy) corresponded to TCP of 93.6% and 97.2%. Overall, the quality of dosimetric reporting was poor; recommended reporting guidelines are presented. CONCLUSIONS: With current typical SRS doses of 12 Gy in 1 fraction, 18 Gy in 3 fractions, and 25 Gy in 5 fractions, 3- to 5-year TCP exceeds 91%. To improve pooled data analyses to optimize treatment outcomes for patients with vestibular schwannoma, future reports of SRS should include complete dosimetric details with well-defined tumor control and toxicity endpoints.

Single- and Multi-Fraction Stereotactic Radiosurgery Dose Tolerances of the Optic Pathways.

PURPOSE: Dosimetric and clinical predictors of radiation-induced optic nerve/chiasm neuropathy (RION) after single-fraction stereotactic radiosurgery (SRS) or hypofractionated (2-5 fractions) radiosurgery (fSRS) were analyzed from pooled data that were extracted from published reports (PubMed indexed from 1990 to June 2015). This study was undertaken as part of the American Association of Physicists in Medicine Working Group on Stereotactic Body Radiotherapy, investigating normal tissue complication probability (NTCP) after hypofractionated radiation. METHODS AND MATERIALS: Eligible studies described dose delivered to optic nerve/chiasm and provided crude or actuarial toxicity risks, with visual endpoints (ie, loss of visual acuity, alterations in visual fields, and/or blindness/complete vision loss). Studies of patients with optic nerve sheath tumors, optic nerve gliomas, or ocular/uveal melanoma were excluded to obviate direct tumor effects on visual outcomes, as were studies not specifying causes of vision loss (ie, tumor progression vs RION). RESULTS: Thirty-four studies (1578 patients) were analyzed. Histologies included pituitary adenoma, cavernous sinus meningioma, craniopharyngioma, and malignant skull base tumors. Prior resection (76% of patients) did not correlate with RION risk (P = .66). Prior irradiation (6% of patients) was associated with a crude 10-fold increased RION risk versus no prior radiation therapy. In patients with no prior radiation therapy receiving SRS/fSRS in 1-5 fractions, optic apparatus maximum point doses resulting in <1% RION risks include 12 Gy in 1 fraction (which is greater than our recommendation of 10 Gy in 1 fraction), 20 Gy in 3 fractions, and 25 Gy in 5 fractions. Omitting multi-fraction data (and thereby eliminating uncertainties associated with dose conversions), a single-fraction dose of 10 Gy was associated with a 1% RION risk. Insufficient details precluded modeling of NTCP risks after prior radiation therapy. CONCLUSIONS: Optic apparatus NTCP and tolerance doses after single- and multi-fraction stereotactic radiosurgery are presented. Additional standardized dosimetric and toxicity reporting is needed to facilitate future pooled analyses and better define RION NTCP after SRS/fSRS.

Local Control After Stereotactic Body Radiation Therapy for Stage I Non-Small Cell Lung Cancer.

PURPOSE: Numerous dose and fractionation schedules have been used to treat medically inoperable stage I non-small cell lung cancer (NSCLC) with stereotactic body radiation therapy (SBRT) or stereotactic ablative radiation therapy. We evaluated published experiences with SBRT to determine local control (LC) rates as a function of SBRT dose. METHODS AND MATERIALS: One hundred sixty published articles reporting LC rates after SBRT for stage I NSCLC were identified. Quality of the series was assessed by evaluating the number of patients in the study, homogeneity of the dose regimen, length of follow-up time, and reporting of LC. Clinical data including 1, 2, 3, and 5-year tumor control probabilities for stages T1, T2, and combined T1 and T2 as a function of the biological effective dose were fitted to the linear quadratic, universal survival curve, and regrowth models. RESULTS: Forty-six studies met inclusion criteria. As measured by the goodness of fit χ2/ndf, with ndf as the number of degrees of freedom, none of the models were ideal fits for the data. Of the 3 models, the regrowth model provides the best fit to the clinical data. For the regrowth model, the fitting yielded an α-to-ß ratio of approximately 25 Gy for T1 tumors, 19 Gy for T2 tumors, and 21 Gy for T1 and T2 combined. To achieve the maximal LC rate, the predicted physical dose schemes when prescribed at the periphery of the planning target volume are 43 ± 1 Gy in 3 fractions, 47 ± 1 Gy in 4 fractions, and 50 ± 1 Gy in 5 fractions for combined T1 and T2 tumors. CONCLUSIONS: Early-stage NSCLC is radioresponsive when treated with SBRT or stereotactic ablative radiation therapy. A steep dose-response relationship exists with high rates of durable LC when physical doses of 43-50 Gy are delivered in 3 to 5 fractions.

An artificial intelligence framework integrating longitudinal electronic health records with real-world data enables continuous pan-cancer prognostication.

Despite widespread adoption of electronic health records (EHRs), most hospitals are not ready to implement data science research in the clinical pipelines. Here, we develop MEDomics, a continuously learning infrastructure through which multimodal health data are systematically organized and data quality is assessed with the goal of applying artificial intelligence for individual prognosis. Using this framework, currently composed of thousands of individuals with cancer and millions of data points over a decade of data recording, we demonstrate prognostic utility of this framework in oncology. As proof of concept, we report an analysis using this infrastructure, which identified the Framingham risk score to be robustly associated with mortality among individuals with early-stage and advanced-stage cancer, a potentially actionable finding from a real-world cohort of individuals with cancer. Finally, we show how natural language processing (NLP) of medical notes could be used to continuously update estimates of prognosis as a given individual's disease course unfolds.

NRG Oncology Survey on Practice and Technology Use in SRT and SBRT Delivery.

PURPOSE: To assess stereotactic radiotherapy (SRT)/stereotactic body radiotherapy (SBRT) practices by polling clinics participating in multi-institutional clinical trials. METHODS: The NRG Oncology Medical Physics Subcommittee distributed a survey consisting of 23 questions, which covered general technologies, policies, and procedures used in the Radiation Oncology field for the delivery of SRT/SBRT (9 questions), and site-specific questions for brain SRT, lung SBRT, and prostate SBRT (14 questions). Surveys were distributed to 1,996 radiotherapy institutions included on the membership rosters of the five National Clinical Trials Network (NCTN) groups. Patient setup, motion management, target localization, prescriptions, and treatment delivery technique data were reported back by 568 institutions (28%). RESULTS: 97.5% of respondents treat lung SBRT patients, 77.0% perform brain SRT, and 29.1% deliver prostate SBRT. 48.8% of clinics require a physicist present for every fraction of SBRT, 18.5% require a physicist present for the initial SBRT fraction only, and 14.9% require a physicist present for the entire first fraction, including set-up approval for all subsequent fractions. 55.3% require physician approval for all fractions, and 86.7% do not reposition without x-ray imaging. For brain SRT, most institutions (83.9%) use a planning target volume (PTV) margin of 2 mm or less. Lung SBRT PTV margins of 3 mm or more are used in 80.6% of clinics. Volumetric modulated arc therapy (VMAT) is the dominant delivery method in 62.8% of SRT treatments, 70.9% of lung SBRT, and 68.3% of prostate SBRT. CONCLUSION: This report characterizes SRT/SBRT practices in radiotherapy clinics participating in clinical trials. Data made available here allows the radiotherapy community to compare their practice with that of other clinics, determine what is achievable, and assess areas for improvement.

Targeted transfer learning to improve performance in small medical physics datasets.

PURPOSE: To perform an in-depth evaluation of current state of the art techniques in training neural networks to identify appropriate approaches in small datasets. METHOD: In total, 112,120 frontal-view X-ray images from the NIH ChestXray14 dataset were used in our analysis. Two tasks were studied: unbalanced multi-label classification of 14 diseases, and binary classification of pneumonia vs non-pneumonia. All datasets were randomly split into training, validation, and testing (70%, 10%, and 20%). Two popular convolution neural networks (CNNs), DensNet121 and ResNet50, were trained using PyTorch. We performed several experiments to test: (a) whether transfer learning using pretrained networks on ImageNet are of value to medical imaging/physics tasks (e.g., predicting toxicity from radiographic images after training on images from the internet), (b) whether using pretrained networks trained on problems that are similar to the target task helps transfer learning (e.g., using X-ray pretrained networks for X-ray target tasks), (c) whether freeze deep layers or change all weights provides an optimal transfer learning strategy, (d) the best strategy for the learning rate policy, and (e) what quantity of data is needed in order to appropriately deploy these various strategies (N = 50 to N = 77 880). RESULTS: In the multi-label problem, DensNet121 needed at least 1600 patients to be comparable to, and 10 000 to outperform, radiomics-based logistic regression. In classifying pneumonia vs non-pneumonia, both CNN and radiomics-based methods performed poorly when N < 2000. For small datasets ( < 2000), however, a significant boost in performance (>15% increase on AUC) comes from a good selection of the transfer learning dataset, dropout, cycling learning rate, and freezing and unfreezing of deep layers as training progresses. In contrast, if sufficient data are available (>35 000), little or no tweaking is needed to obtain impressive performance. While transfer learning using X-ray images from other anatomical sites improves performance, we also observed a similar boost by using pretrained networks from ImageNet. Having source images from the same anatomical site, however, outperforms every other methodology, by up to 15%. In this case, DL models can be trained with as little as N = 50. CONCLUSIONS: While training DL models in small datasets (N < 2000) is challenging, no tweaking is necessary for bigger datasets (N > 35 000). Using transfer learning with images from the same anatomical site can yield remarkable performance in new tasks with as few as N = 50. Surprisingly, we did not find any advantage for using images from other anatomical sites over networks that have been trained using ImageNet. This indicates that features learned may not be as general as currently believed, and performance decays rapidly even by just changing the anatomical site of the images.

DoseGAN: a generative adversarial network for synthetic dose prediction using attention-gated discrimination and generation.

Deep learning algorithms have recently been developed that utilize patient anatomy and raw imaging information to predict radiation dose, as a means to increase treatment planning efficiency and improve radiotherapy plan quality. Current state-of-the-art techniques rely on convolutional neural networks (CNNs) that use pixel-to-pixel loss to update network parameters. However, stereotactic body radiotherapy (SBRT) dose is often heterogeneous, making it difficult to model using pixel-level loss. Generative adversarial networks (GANs) utilize adversarial learning that incorporates image-level loss and is better suited to learn from heterogeneous labels. However, GANs are difficult to train and rely on compromised architectures to facilitate convergence. This study suggests an attention-gated generative adversarial network (DoseGAN) to improve learning, increase model complexity, and reduce network redundancy by focusing on relevant anatomy. DoseGAN was compared to alternative state-of-the-art dose prediction algorithms using heterogeneity index, conformity index, and various dosimetric parameters. All algorithms were trained, validated, and tested using 141 prostate SBRT patients. DoseGAN was able to predict more realistic volumetric dosimetry compared to all other algorithms and achieved statistically significant improvement compared to all alternative algorithms for the V100 and V120 of the PTV, V60 of the rectum, and heterogeneity index.

Expert-augmented machine learning.

Machine learning is proving invaluable across disciplines. However, its success is often limited by the quality and quantity of available data, while its adoption is limited by the level of trust afforded by given models. Human vs. machine performance is commonly compared empirically to decide whether a certain task should be performed by a computer or an expert. In reality, the optimal learning strategy may involve combining the complementary strengths of humans and machines. Here, we present expert-augmented machine learning (EAML), an automated method that guides the extraction of expert knowledge and its integration into machine-learned models. We used a large dataset of intensive-care patient data to derive 126 decision rules that predict hospital mortality. Using an online platform, we asked 15 clinicians to assess the relative risk of the subpopulation defined by each rule compared to the total sample. We compared the clinician-assessed risk to the empirical risk and found that, while clinicians agreed with the data in most cases, there were notable exceptions where they overestimated or underestimated the true risk. Studying the rules with greatest disagreement, we identified problems with the training data, including one miscoded variable and one hidden confounder. Filtering the rules based on the extent of disagreement between clinician-assessed risk and empirical risk, we improved performance on out-of-sample data and were able to train with less data. EAML provides a platform for automated creation of problem-specific priors, which help build robust and dependable machine-learning models in critical applications.

Attention-Aware Discrimination for MR-to-CT Image Translation Using Cycle-Consistent Generative Adversarial Networks.

PURPOSE: To suggest an attention-aware, cycle-consistent generative adversarial network (A-CycleGAN) enhanced with variational autoencoding (VAE) as a superior alternative to current state-of-the-art MR-to-CT image translation methods. MATERIALS AND METHODS: An attention-gating mechanism is incorporated into a discriminator network to encourage a more parsimonious use of network parameters, whereas VAE enhancement enables deeper discrimination architectures without inhibiting model convergence. Findings from 60 patients with head, neck, and brain cancer were used to train and validate A-CycleGAN, and findings from 30 patients were used for the holdout test set and were used to report final evaluation metric results using mean absolute error (MAE) and peak signal-to-noise ratio (PSNR). RESULTS: A-CycleGAN achieved superior results compared with U-Net, a generative adversarial network (GAN), and a cycle-consistent GAN. The A-CycleGAN averages, 95% confidence intervals (CIs), and Wilcoxon signed-rank two-sided test statistics are shown for MAE (19.61 [95% CI: 18.83, 20.39], P = .0104), structure similarity index metric (0.778 [95% CI: 0.758, 0.798], P = .0495), and PSNR (62.35 [95% CI: 61.80, 62.90], P = .0571). CONCLUSION: A-CycleGANs were a superior alternative to state-of-the-art MR-to-CT image translation methods.© RSNA, 2020.

Photons or protons for reirradiation in (non-)small cell lung cancer: Results of the multicentric ROCOCO in silico study.

OBJECTIVE: Locally recurrent disease is of increasing concern in (non-)small cell lung cancer [(N)SCLC] patients. Local reirradiation with photons or particles may be of benefit to these patients. In this multicentre in silico trial performed within the Radiation Oncology Collaborative Comparison (ROCOCO) consortium, the doses to the target volumes and organs at risk (OARs) were compared when using several photon and proton techniques in patients with recurrent localised lung cancer scheduled to undergo reirradiation. METHODS: 24 consecutive patients with a second primary (N)SCLC or recurrent disease after curative-intent, standard fractionated radio(chemo)therapy were included in this study. The target volumes and OARs were centrally contoured and distributed to the participating ROCOCO sites. Remaining doses to the OARs were calculated on an individual patient's basis. Treatment planning was performed by the participating site using the clinical treatment planning system and associated beam characteristics. RESULTS: Treatment plans for all modalities (five photon and two proton plans per patient) were available for 22 patients (N = 154 plans). 3D-conformal photon therapy and double-scattered proton therapy delivered significantly lower doses to the target volumes. The highly conformal techniques, i.e., intensity modulated radiation therapy (IMRT), volumetric modulated arc therapy (VMAT), CyberKnife, TomoTherapy and intensity-modulated proton therapy (IMPT), reached the highest doses in the target volumes. Of these, IMPT was able to statistically significantly decrease the radiation doses to the OARs. CONCLUSION: Highly conformal photon and proton beam techniques enable high-dose reirradiation of the target volume. They, however, significantly differ in the dose deposited in the OARs. The therapeutic options, i.e., reirradiation or systemic therapy, need to be carefully weighed and discussed with the patients. ADVANCES IN KNOWLEDGE: Highly conformal photon and proton beam techniques enable high-dose reirradiation of the target volume. In light of the abilities of the various highly conformal techniques to spare specific OARs, the therapeutic options need to be carefully weighed and patients included in the decision-making process.

Integrated models incorporating radiologic and radiomic features predict meningioma grade, local failure, and overall survival.

BACKGROUND: We investigated prognostic models based on clinical, radiologic, and radiomic feature to preoperatively identify meningiomas at risk for poor outcomes. METHODS: Retrospective review was performed for 303 patients who underwent resection of 314 meningiomas (57% World Health Organization grade I, 35% grade II, and 8% grade III) at two independent institutions, which comprised primary and external datasets. For each patient in the primary dataset, 16 radiologic and 172 radiomic features were extracted from preoperative magnetic resonance images, and prognostic features for grade, local failure (LF) or overall survival (OS) were identified using the Kaplan-Meier method, log-rank tests and recursive partitioning analysis. Regressions and random forests were used to generate and test prognostic models, which were validated using the external dataset. RESULTS: Multivariate analysis revealed that apparent diffusion coefficient hypointensity (HR 5.56, 95% CI 2.01-16.7, P = .002) was associated with high grade meningioma, and low sphericity was associated both with increased LF (HR 2.0, 95% CI 1.1-3.5, P = .02) and worse OS (HR 2.94, 95% CI 1.47-5.56, P = .002). Both radiologic and radiomic predictors of adverse meningioma outcomes were significantly associated with molecular markers of aggressive meningioma biology, such as somatic mutation burden, DNA methylation status, and FOXM1 expression. Integrated prognostic models combining clinical, radiologic, and radiomic features demonstrated improved accuracy for meningioma grade, LF, and OS (area under the curve 0.78, 0.75, and 0.78, respectively) compared to models based on clinical features alone. CONCLUSIONS: Preoperative radiologic and radiomic features such as apparent diffusion coefficient and sphericity can predict tumor grade, LF, and OS in patients with meningioma.

Building more accurate decision trees with the additive tree.

The expansion of machine learning to high-stakes application domains such as medicine, finance, and criminal justice, where making informed decisions requires clear understanding of the model, has increased the interest in interpretable machine learning. The widely used Classification and Regression Trees (CART) have played a major role in health sciences, due to their simple and intuitive explanation of predictions. Ensemble methods like gradient boosting can improve the accuracy of decision trees, but at the expense of the interpretability of the generated model. Additive models, such as those produced by gradient boosting, and full interaction models, such as CART, have been investigated largely in isolation. We show that these models exist along a spectrum, revealing previously unseen connections between these approaches. This paper introduces a rigorous formalization for the additive tree, an empirically validated learning technique for creating a single decision tree, and shows that this method can produce models equivalent to CART or gradient boosted stumps at the extremes by varying a single parameter. Although the additive tree is designed primarily to provide both the model interpretability and predictive performance needed for high-stakes applications like medicine, it also can produce decision trees represented by hybrid models between CART and boosted stumps that can outperform either of these approaches.

Attention-enabled 3D boosted convolutional neural networks for semantic CT segmentation using deep supervision.

A deeply supervised attention-enabled boosted convolutional neural network (DAB-CNN) is presented as a superior alternative to current state-of-the-art convolutional neural networks (CNNs) for semantic CT segmentation. Spatial attention gates (AGs) were incorporated into a novel 3D cascaded CNN framework to prioritize relevant anatomy and suppress redundancies within the network. Due to the complexity and size of the network, incremental channel boosting was used to decrease memory usage and facilitate model convergence. Deep supervision was used to encourage semantically meaningful deep features and mitigate local minima traps during training. The accuracy of DAB-CNN is compared to seven architectures: a variation of U-Net (UNet), attention-enabled U-Net (A-UNet), boosted U-Net (B-UNet), deeply-supervised U-Net (D-UNet), U-Net with ResNeXt blocks (ResNeXt), life-long learning segmentation CNN (LL-CNN), and deeply supervised attention-enabled U-Net (DA-UNet). The accuracy of each method was assessed based on Dice score compared to manually delineated contours as the gold standard. One hundred and twenty patients who had definitive prostate radiotherapy were used in this study. Training, validation, and testing followed Kaggle competition rules, with 80 patients used for training, 20 patients used for internal validation, and 20 test patients used to report final accuracies. Comparator p -values indicate that DAB-CNN achieved significantly superior Dice scores than all alternative algorithms for the prostate, rectum, and penile bulb. This study demonstrated that attention-enabled boosted convolutional neural networks (CNNs) using deep supervision are capable of achieving superior prediction accuracy compared to current state-of-the-art automatic segmentation methods.