Search

PurposeMachine 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.MethodsWe 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).ResultsRandom 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.ConclusionRandom 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.

The BRASA patient decision aid (BRASA-PtDA) facilitates shared decision making for breast cancer patients (BCPs) facing a radiotherapy treatment decision. During evaluations, patients indicated the wish for quantitative information on side effects. Therefore, this study assessed BCPs opinion on which and how information on side effects should be incorporated in the BRASA-PtDA. A workshop was organized with BCPs (n = 9), researchers (n = 5) and clinicians (n = 3). Subsequently, a survey was sent to BCPs (n = 744) investigating the generalisability of the workshop findings, and posing additional questions. The survey entailed multiple choice questions on quality of life themes, the use of a decision aid and risk communication. The workshop revealed BCPs wish for a layered, all encompassing information system. Information on the impact of side effects on daily life was preferred above the risk of these side effects. The survey revealed that important quality of life (QoL) themes were having energy (81%; n = 605), arm function (61%; n = 452), pain (55%; n = 410). Despite the focus on qualitative effects in the workshop, 89% of the survey respondents also wanted to be informed on individualized risks of side effects. 54% Of the survey respondents had never heard of a PtDA. BCPs preferred information on the impact of side effects, but also their individualized risks on side effects. Most important QoL themes were having enough energy, arm function and pain. Consequently, the BRASA-PtDA should be reshaped, starting with quality of life themes, rather than side effects. • Having enough energy is the most important QoL theme for patients. • The practical impact of side effects on daily life are more important than personal risks. • Half of the former patients did not know what a decision aid was. • Patients stress the importance of a layered information system. [ABSTRACT FROM AUTHOR]

Purpose: Personalized medicine is expected to yield improved health outcomes. Data mining over massive volumes of patients' clinical data is an appealing, low‐cost and noninvasive approach toward personalization. Machine learning algorithms could be trained over clinical "big data" to build prediction models for personalized therapy. To reach this goal, a scalable "big data" architecture for the medical domain becomes essential, based on data standardization to transform clinical data into FAIR (Findable, Accessible, Interoperable and Reusable) data. Using Ontologies and Semantic Web technologies, we attempt to reach mentioned goal. Methods: We developed an ontology to be used in the field of radiation oncology to map clinical data from relational databases. We combined ontology with semantic Web techniques to publish mapped data and easily query them using SPARQL. Results: The Radiation Oncology Ontology (ROO) contains 1,183 classes and 211 properties between classes to represent clinical data (and their relationships) in the radiation oncology domain following FAIR principles. We combined the ontology with Semantic Web technologies showing how to efficiently and easily integrate and query data from different (relational database) sources without a priori knowledge of their structures. Discussion: When clinical FAIR data sources are combined (linked data) using mentioned technologies, new relationships between entities are created and discovered, representing a dynamic body of knowledge that is continuously accessible and increasing. [ABSTRACT FROM AUTHOR]

Purpose/Objective:Chemo-radiotherapy is an established primary curative treatment for anal cancer, but clinically equal rationale for different target doses exists. If joint preferences (physician and patient) are used to determine acceptable tradeoffs in radiotherapy treatment planning, multiple dose plans must be simultaneously explored. We quantified the degree to which different toxicity priorities might be incorporated into treatment plan selection, to elucidate the feasible decision space for shared decision making in anal cancer radiotherapy. Material and methods:Retrospective plans were generated for 22 anal cancer patients. Multi-criteria optimization handles dynamically changing priorities between clinical objectives while meeting fixed clinical constraints. Four unique dose distributions were designed to represent a wide span of clinically relevant objectives: high-dose preference (60.2 Gy tumor boost and 50.4 Gy to elective nodes with physician-defined order of priorities), low-dose preference (53.75 Gy tumor boost, 45 Gy to elective nodes, physician-defined priorities), bowel sparing preference (lower dose levels and priority for bowel avoidance) and bladder sparing preference (lower dose levels and priority for bladder avoidance). Results:Plans satisfied constraints for target coverage. A senior oncologist approved a random subset of plans for quality assurance. Compared to a high-dose preference, bowel sparing was clinically meaningful at the lower prescribed dose [median change in V45Gy: 234 cm3; inter-quartile range (66; 247);p < .01] and for a bowel sparing preference [median change in V45Gy: 281 cm3; (73; 488);p < .01]. Compared to a high-dose preference, bladder sparing was clinically meaningful at the lower prescribed dose [median change in V35Gy: 13.7%-points; (0.3; 30.6);p < .01] and for a bladder sparing preference [median change in V35Gy: 30.3%-points; (12.4; 43.1);p < .01]. Conclusions:There is decision space available in anal cancer radiotherapy to incorporate preferences, although tradeoffs are highly patient-dependent. This study demonstrates that preference-informed dose planning is feasible for clinical studies utilizing shared decision making. [ABSTRACT FROM AUTHOR]

• A CT-derived radiomics model can predict OS in patients with a glioblastoma. • Discrimination based on the combined clinical and radiomics model was comparable to previous MRI-based models. • Qualitatively high-level datasets will support further model development. Glioblastoma (GBM) is the most common malignant primary brain tumour which has, despite extensive treatment, a median overall survival of 15 months. Radiomics is the high-throughput extraction of large amounts of image features from radiographic images, which allows capturing the tumour phenotype in 3D and in a non-invasive way. In this study we assess the prognostic value of CT radiomics for overall survival in patients with a GBM. Clinical data and pre-treatment CT images were obtained from 218 patients diagnosed with a GBM via biopsy who underwent radiotherapy +/− temozolomide between 2004 and 2015 treated at three independent institutes (n = 93, 62 and 63). A clinical prognostic score (CPS), a simple radiomics model consisting of volume based score (VPS), a complex radiomics prognostic score (RPS) and a combined clinical and radiomics (C + R)PS model were developed. The population was divided into three risk groups for each prognostic score and respective Kaplan–Meier curves were generated. Patient characteristics were broadly comparable. Clinically significant differences were observed with regards to radiation dose, tumour volume and performance status between datasets. Image acquisition parameters differed between institutes. The cross-validated c-indices were moderately discriminative and for the CPS ranged from 0.63 to 0.65; the VPS c-indices ranged between 0.52 and 0.61; the RPS c-indices ranged from 0.57 to 0.64 and the combined clinical and radiomics model resulted in c-indices of 0.59–0.71. In this study clinical and CT radiomics features were used to predict OS in GBM. Discrimination between low-, middle- and high-risk patients based on the combined clinical and radiomics model was comparable to previous MRI-based models. [ABSTRACT FROM AUTHOR]

• CT and radiation dose can play synergistically with each other to model. • Deep learning methods have higher predictive power than traditional methods. • The predictive logic of deep learning can be inferred by activation maps. • Simple tuning method can be used in deep learning model to adapt to different data. To develop a deep learning model that combines CT and radiation dose (RD) images to predict the occurrence of radiation pneumonitis (RP) in lung cancer patients who received radical (chemo)radiotherapy. CT, RD images and clinical parameters were obtained from 314 retrospectively-collected patients (training set) and 35 prospectively-collected patients (test-set-1) who were diagnosed with lung cancer and received radical radiotherapy in the dose range of 50 Gy and 70 Gy. Another 194 (60 Gy group, test-set-2) and 158 (74 Gy group, test-set-3) patients from the clinical trial RTOG 0617 were used for external validation. A ResNet architecture was used to develop a prediction model that combines CT and RD features. Thereafter, the CT and RD weights were adjusted by using 40 patients from test-set-2 or 3 to accommodate cohorts with different clinical settings or dose delivery patterns. Visual interpretation was implemented using a gradient-weighted class activation map (grad-CAM) to observe the area of model attention during the prediction process. To improve the usability, ready-to-use online software was developed. The discriminative ability of a baseline trained model had an AUC of 0.83 for test-set-1, 0.55 for test-set-2, and 0.63 for test-set-3. After adjusting CT and RD weights of the model using a subset of the RTOG-0617 subjects, the discriminatory power of test-set-2 and 3 improved to AUC 0.65 and AUC 0.70, respectively. Grad-CAM showed the regions of interest to the model that contribute to the prediction of RP. A novel deep learning approach combining CT and RD images can effectively and accurately predict the occurrence of RP, and this model can be adjusted easily to fit new cohorts. [ABSTRACT FROM AUTHOR]