Search

Purpose: Organ‐at‐risk contouring is still a bottleneck in radiotherapy, with many deep learning methods falling short of promised results when evaluated on clinical data. We investigate the accuracy and time‐savings resulting from the use of an interactive‐machine‐learning method for an organ‐at‐risk contouring task. Methods: We implement an open‐source interactive‐machine‐learning software application that facilitates corrective‐annotation for deep‐learning generated contours on X‐ray CT images. A trained‐physician contoured 933 hearts using our software by delineating the first image, starting model training, and then correcting the model predictions for all subsequent images. These corrections were added into the training data, which was used for continuously training the assisting model. From the 933 hearts, the same physician also contoured the first 10 and last 10 in Eclipse (Varian) to enable comparison in terms of accuracy and duration. Results: We find strong agreement with manual delineations, with a dice score of 0.95. The annotations created using corrective‐annotation also take less time to create as more images are annotated, resulting in substantial time savings compared to manual methods. After 923 images had been delineated, hearts took 2 min and 2 s to delineate on average, which includes time to evaluate the initial model prediction and assign the needed corrections, compared to 7 min and 1 s when delineating manually. Conclusions: Our experiment demonstrates that interactive‐machine‐learning with corrective‐annotation provides a fast and accessible way for non computer‐scientists to train deep‐learning models to segment their own structures of interest as part of routine clinical workflows. [ABSTRACT FROM AUTHOR]

Introduction: The exact dependence of biological effect on dose and linear energy transfer (LET) in human tissue when delivering proton therapy is unknown. In this study, we propose a framework for measuring this dependency using multi‐modal image‐based assays with deformable registrations within imaging sessions and across time. Materials and Methods: 3T MRI scans were prospectively collected from 6 pediatric brain cancer patients before they underwent proton therapy treatment, and every 3 months for a year after treatment. Scans included T1‐weighted with contrast enhancement (T1), T2‐FLAIR (T2) and fractional anisotropy (FA) images. In addition, the planning CT, dose distributions and Monte Carlo‐calculated LET distributions were collected. A multi‐modal deformable image registration framework was used to create a dataset of dose, LET and imaging intensities at baseline and follow‐up on a voxel‐by‐voxel basis. We modelled the biological effect of dose and LET from proton therapy using imaging changes over time as a surrogate for biological effect. We investigated various models to show the feasibility of the framework to model imaging changes. To account for interpatient and intrapatient variations, we used a nested generalized linear mixed regression model. The models were applied to predict imaging changes over time as a function of dose and LET for each modality. Results: Using the nested models to predict imaging changes, we saw a decrease in the FA signal as a function of dose; however, the signal increased with increasing LET. Similarly, we saw an increase in T2 signal as a function of dose, but a decrease in signal with LET. We saw no changes in T1 voxel values as a function of either dose or LET. Conclusions: The imaging changes could successfully model biological effect as a function of dose and LET using our proposed framework. Due to the low number of patients, the imaging changes observed for FA and T2 scans were not marked enough to draw any firm conclusions. [ABSTRACT FROM AUTHOR]

Background: Isolated regional recurrences following head‐neck squamous‐cell carcinomas (HNSCC) are often accessible for curatively intended salvage treatment. Factors prognostic for outcome were investigated in a large cohort of HNSCC patients. Methods: In total, 1811 patients receiving curatively intended radiotherapy from 2007 to 2017 were reviewed and isolated cervical nodal recurrences were identified. Factors associated with survival and second recurrence were investigated using univariate and multivariate analyses. Results: Isolated regional recurrence was seen in 95/1811 (5.2%) patients. Eighty of 95 patients (84%) received salvage surgery. Two‐year survival after isolated regional recurrence was 40%. Overall survival (OS) and time to second recurrence were associated with resection status of the salvage surgery and presence of extranodal spread (ENS), while p16‐positive oropharyngeal squamous‐cell carcinoma (OPSCC) was associated with better OS. Conclusion: Long‐term survival after regional recurrence in HNSCC is possible. p16‐positive OPSCC, complete salvage surgery, and lack of ENS are associated with better outcome. [ABSTRACT FROM AUTHOR]

Purpose: Radiation therapy (RT) is prescribed for curative and palliative treatment for around 50% of patients with solid tumors. Radiation‐induced toxicities of healthy organs accompany many RTs and represent one of the main limiting factors during dose delivery. The existing RT planning solutions generally discard spatial dose distribution information and lose the ability to recognize radiosensitive regions of healthy organs potentially linked to toxicity manifestation. This study proposes a universal deep learning‐based algorithm for recognitions of consistent dose patterns and generation of toxicity risk maps for the abdominal area. Methods: We investigated whether convolutional neural networks (CNNs) can automatically associate abdominal computed tomography (CT) images and RT dose plans with post‐RT toxicities without being provided segmentation of abdominal organs. The CNNs were also applied to study RT plans, where doses at specific anatomical regions were reduced/increased, with the aim to pinpoint critical regions sparing of which significantly reduces toxicity risks. The obtained risk maps were computed for individual anatomical regions inside the liver and statistically compared to the existing clinical studies. Results: A database of 122 liver stereotactic body RT (SBRT) executed at Stanford Hospital from July 2004 and November 2015 was assembled. All patients treated for primary liver cancer, mainly hepatocellular carcinoma and cholangiocarcinoma, with complete follow‐ups were extracted from the database. The SBRT treatment doses ranged from 26 to 50 Gy delivered in 1–5 fractions for primary liver cancer. The patients were followed up for 1–68 months depending on the survival time. The CNNs were trained to recognize acute and late grade 3+ biliary stricture/obstruction, hepatic failure or decompensation, hepatobiliary infection, liver function test (LFT) elevation or/and portal vein thrombosis, named for convenience hepatobiliary (HB) toxicities. The toxicity prediction accuracy was of 0.73 measured in terms of the area under the receiving operator characteristic curve. Significantly higher risk scores (P < 0.05) of HB toxicity manifestation were associated with irradiation for the hepatobiliary tract in comparison to the risk scores for liver segments I–VIII and portal vein. This observation is in strong agreement with anatomical and clinical expectations. Conclusion: In this work, we proposed and validated a universal deep learning‐based solution for the identification of radiosensitive anatomical regions. Without any prior anatomical knowledge, CNNs automatically recognized the importance of hepatobiliary tract sparing during liver SBRT. [ABSTRACT FROM AUTHOR]

Radiation oncology, a major treatment modality in the care of patients with malignant disease, is a technology‐ and computer‐intensive medical specialty. As such, it should lend itself ideally to data science methods, where computer science, statistics, and clinical knowledge are combined to advance state‐of‐the‐art care. Nevertheless, data science methods in radiation oncology research are still in their infancy and successful applications leading to improved patient care remain scarce. Here, we discuss data interoperability issues within and across organizational boundaries that hamper the introduction of big data and data science techniques in radiation oncology. At the semantic level, creating common underlying models and codification of the data, including the use of data elements with standardized definitions, an ontology, remains a work in progress. Methodological issues in data science and in the use of large population‐based health data registries are identified. We show that data science methods and big data cannot replace randomized clinical trials in comparative effectiveness research by reviewing a series of instances where the outcomes of big data analyses and randomized trials are at odds. We also discuss the modern wave of machine learning and artificial intelligence as represented by deep learning and convolutional neural networks. Finally, we identify promising research avenues and remain optimistic that the data sources in radiation oncology can be linked to yield important insights in the near future. We argue that data science will be a valuable complement to, but not a replacement of, the traditional hypothesis‐driven translational research chain and the randomized clinical trials that form the backbone of evidence‐based medicine. [ABSTRACT FROM AUTHOR]

Purpose: Current radiation therapy planning uses a set of defined dose–volume constraints to ensure a specified level of tumor coverage while constraining the dose distribution in the organs at risk. Such constraints are aggregated, population‐based quantities that do not adequately consider patient‐specific risk factors. Furthermore, these constraints are calculated for each organ independently and it is therefore not guaranteed that the optimal trade‐off between organs is achieved. We introduce a novel radiotherapy planning approach where a patient‐specific all‐cause mortality risk is minimized using inverse plan optimization. As illustration of concept, our outcome risk model incorporates patient age, sex, cardiac risk factor (CRF), and smoking. Methods and materials: We retrospectively analyzed a left‐sided breast cancer case and a Hodgkin's lymphoma case, both clinically treated with three‐dimensional conformal radiotherapy (3D‐CRT). Our objective function for inverse plan optimization was an equally weighted summation of risk models for cancer recurrence and mortality from radiation‐induced coronary heart disease and secondary lung and breast cancers incorporating patient age, sex, CRF, and smoking. We allowed the optimization algorithm to choose from a large set of gantry angles. The optimization task was to choose beams and optimize monitor units (MUs) so that overall survival was maximized (and the total risk of cancer recurrence and mortality from radiation‐induced causes were minimized). The sensitivity analysis was performed in the lymphoma case by changing the tumor control probability model from using mean dose (Model 1) to using generalized equivalent uniform dose (Model 2). Results: For the breast case in this study, the 3D‐CRT clinical plan used eight beams while the proposed 3D‐CRT outcome‐optimized plan used five beams, reducing the total risk — summation of the risks of recurrence and secondary disease mortality — from 3% to 2%. The mean doses to clinical target volume (CTV) and internal mammary nodes (IMN) were increased in the outcome‐optimized plan by 1.9 and 1.8 Gy, respectively. For the Hodgkin's lymphoma case, the clinical 3D‐CRT plan used two beams, while the proposed 3D‐CRT outcome‐optimized plan used three beams, reducing the total risk by 6% (from 16% to 10%). Using either of the two tumor control models for the lymphoma case resulted in outcome‐optimized plans where tumor control was compensated at the cost of saving organs at risk. However, the impact of sensitivity to models was comparatively large. Using Model 1 resulted in a reduction in mean target dose by 15.2 vs 7.1 Gy for Model 2. In all cases, the chosen beams in outcome‐optimized plans were different from clinically used beams. Conclusions: The proposed optimization strategy, supplanting dosimetric objectives with comprehensive individual risk estimates, has the potential to yield improved outcomes in terms of reduced mortality risk in cancer patients treated with radiotherapy. The approach is, however, currently limited by gaps in knowledge about the effect of compromising dose to part of the target, for example, in order to spare cardiac structures. [ABSTRACT FROM AUTHOR]

Background The prognostic value of 18F-Fludeoxyglucose (FDG) uptake could be due to its association with already known clinical risk factors. Methods Correlation between FDG uptake metrics and other known risk factors from 287 patients were analyzed. Time to any failure was analyzed using Cox proportional hazards model stratified by tumor subsite. The resulting multivariate prognostic model was used to generate a table of 2-year freedom from failure estimates with confidence intervals (CIs). Results Increasing values of standardized uptake value maximum (SUVmax) correlated with other known risk factors. The reduced Cox model included SUVmax (hazard ratio [HR] = 1.34), cisplatin (HR = 0.37), smoking status (HR = 1.49), and gross target volume (GTV; HR = 1.74) as significant prognostic factors. Including SUVmax in the model changed the 2-year failure estimate by more than 10% for a quarter of the patients (23%). Conclusion FDG uptake retains statistical significance in a multivariate analysis and has clinically relevant prognostic impact. We developed a prognostic model for risk stratification of patients in a clinical setting. © 2014 Wiley Periodicals, Inc. Head Neck 37: 1274-1281, 2015 [ABSTRACT FROM AUTHOR]

Purpose: To demonstrate a data-driven dose-painting strategy based on the spatial distribution of recurrences in previously treated patients. The result is a quantitative way to define a dose prescription function, optimizing the predicted local control at constant treatment intensity. A dose planning study using the optimized dose prescription in 20 patients is performed. Methods: Patients treated at our center have five tumor subvolumes from the center of the tumor (PET positive volume) and out delineated. The spatial distribution of 48 failures in patients with complete clinical response after (chemo)radiation is used to derive a model for tumor control probability (TCP). The total TCP is fixed to the clinically observed 70% actuarial TCP at five years. Additionally, the authors match the distribution of failures between the five subvolumes to the observed distribution. The steepness of the dose-response is extracted from the literature and the authors assume 30% and 20% risk of subclinical involvement in the elective volumes. The result is a five-compartment dose response model matching the observed distribution of failures. The model is used to optimize the distribution of dose in individual patients, while keeping the treatment intensity constant and the maximum prescribed dose below 85 Gy. Results: The vast majority of failures occur centrally despite the small volumes of the central regions. Thus, optimizing the dose prescription yields higher doses to the central target volumes and lower doses to the elective volumes. The dose planning study shows that the modified prescription is clinically feasible. The optimized TCP is 89% (range: 82%-91%) as compared to the observed TCP of 70%. Conclusions: The observed distribution of locoregional failures was used to derive an objective, data-driven dose prescription function. The optimized dose is predicted to result in a substantial increase in local control without increasing the predicted risk of toxicity. [ABSTRACT FROM AUTHOR]

BACKGROUND: The authors developed a framework for estimating and comparing the risks of various long-term complications on a common scale and applied it to 3 different techniques for craniospinal irradiation in patients with pediatric medulloblastoma. METHODS: Radiation dose-response parameters related to excess hazard ratios for secondary breast, lung, stomach, and thyroid cancer; heart failure, and myocardial infarction were derived from large published clinical series. Combined with age-specific and sex-specific hazards in the US general population, the dose-response analysis yielded excess hazards of complications for a cancer survivor as a function of attained age. After adjusting for competing risks of death, life years lost (LYL) were estimated based on excess hazard and prognosis of a complication for 3-dimensional conformal radiotherapy (3D CRT), volumetric modulated arc therapy (VMAT), and intensity-modulated proton therapy (IMPT). RESULTS: Lung cancer contributed most to the estimated LYL, followed by myocardial infarction, and stomach cancer. The estimates of breast or thyroid cancer incidence were higher than those for lung and stomach cancer incidence, but LYL were lower because of the relatively good prognosis. Estimated LYL ranged between 1.90 years for 3D CRT to 0.28 years for IMPT. In a paired comparison, IMPT was associated with significantly fewer LYL than both photon techniques. CONCLUSIONS: Estimating the risk of late complications is associated with considerable uncertainty, but including prognosis and attained age at an event to obtain the more informative LYL estimate added relatively little to this uncertainty. Cancer 2012. © 2012 American Cancer Society. [ABSTRACT FROM AUTHOR]

BACKGROUND: A systematic overview and meta-analysis of studies reporting data on hypothyroidism (HT) after radiation therapy was conducted to identify risk factors for development of HT. METHODS: Published studies were identified from the PubMed and Embase databases and by hand-searching published reviews. Studies allowing the extraction of odds ratios (OR) for HT in 1 or more of several candidate clinical risk groups were included. A meta-analysis of the OR for development of HT with or without each of the candidate risk factors was performed. Furthermore, studies allowing the extraction of radiation dose-response data were identified for a meta-analysis of the dose-response curve. RESULTS: Female gender (OR = 1.6; 95% confidence interval [CI], 1.3-1.9; P < .00001), surgery involving the thyroid gland (OR = 8.3; 95% CI, 5.7-12.0; P < .00001), or other neck surgery (OR = 1.7; 95% CI, 1.16-2.42; P = .006) were associated with a higher risk of HT. Caucasians were at higher risk of HT than African Americans (OR = 4.8; 95% CI, 2.8-8.5; P < .00001). The data showed association between lymphangiography and HT but with evidence of publication bias. There was a radiation dose-response relation with a 50% risk of HT at a dose of 45 Gy but with considerable variation in the dose response between studies. Chemotherapy and age were not associated with risk of HT in this analysis. CONCLUSIONS: Several clinical risk factors for HT were identified. The risk of HT increases with increasing radiation dose, but the specific radiation dose response varies between the studies. The most likely cause of this heterogeneity is differences in follow-up between studies. Cancer 2011;. © 2011 American Cancer Society. [ABSTRACT FROM AUTHOR]