Machine Learning in Radiation Oncology: Opportunities, Requirements, and Needs

Feng, Mary; Valdés, Gilmer; Dixit, Nayha; Solberg, Timothy D.

doi:10.3389/fonc.2018.00110

Cited by 93 publications

(84 citation statements)

References 68 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A significant shorter time for each fraction is possible if further improvement in the deformable registration step of the original contours is achieved. However, other alternatives are also possible for the generation of new contours for both tumor and OARs at each fraction, such as the use of atlas based methods [34] or convolutional neural networks [35,36]. The time spent in recontouring and generating a new treatment plan could also be used to acquire additional MR sequences for offline evaluation of treatment response (for instance, diffusion weighted MR [37]).…”

Section: Discussionmentioning

confidence: 99%

Clinical implementation of magnetic resonance imaging guided adaptive radiotherapy for localized prostate cancer

Tetar

Bruynzeel

Lagerwaard

et al. 2019

Physics and Imaging in Radiation Oncology

137

138

View full text Add to dashboard Cite

A B S T R A C TBackground and purpose: Magnetic resonance-guided radiation therapy (MRgRT) has recently become available in clinical practice and is expected to expand significantly in coming years. MRgRT offers marker-less continuous imaging during treatment delivery, use of small clinical target volume (CTV) to planning target volume (PTV) margins, and finally the option to perform daily plan re-optimization. Materials and methods: A total of 140 patients (700 fractions) have been treated with MRgRT and online plan adaptation for localized prostate cancer since early 2016. Clinical workflow for MRgRT of prostate cancer consisted of patient selection, simulation on both MR-and computed tomography (CT) scan, inverse intensitymodulated radiotherapy (IMRT) treatment planning and daily plan re-optimization prior to treatment delivery with partial organs at risk (OAR) recontouring within the first 2 cm outside the PTV. For each adapted plan online patient-specific quality assurance (QA) was performed by means of a secondary Monte Carlo 3D dose calculation and gamma analysis comparison. Patient experiences with MRgRT were assessed using a patientreported outcome questionnaire (PRO-Q) after the last fraction. Results: In 97% of fractions, MRgRT was delivered using the online adapted plan. Intrafractional prostate drifts necessitated 2D-corrections during treatment in approximately 20% of fractions. The average duration of an uneventful fraction of MRgRT was 45 min. showed that MRgRT was generally well tolerated, with disturbing noise sensations being most commonly reported. Conclusions: MRgRT with daily online plan adaptation constitutes an innovative approach for delivering SBRT for prostate cancer and appears to be feasible, although necessitating extended timeslots and logistical challenges.

show abstract

Section: Discussionmentioning

confidence: 99%

Clinical implementation of magnetic resonance imaging guided adaptive radiotherapy for localized prostate cancer

Tetar

Bruynzeel

Lagerwaard

et al. 2019

Physics and Imaging in Radiation Oncology

137

138

View full text Add to dashboard Cite

show abstract

“…Currently, many parts of the medical datasets collected in radiation oncology have unstructured format (e.g., medical notes, images). Understanding how much knowledge can be extracted from them is a challenging and exciting task . Deep learning (DL) is a specific subfield of ML that learns representations from data and it is perfectly suited to handle unstructured data like images or text.…”

Section: Machine Learning Approaches For Radiation Outcome Modelingmentioning

confidence: 99%

Machine learning for radiation outcome modeling and prediction

2020

Self Cite

View full text Add to dashboard Cite

Aims This review paper intends to summarize the application of machine learning to radiotherapy outcome modeling based on structured and un‐structured radiation oncology datasets. Materials and methods The most appropriate machine learning approaches for structured datasets in terms of accuracy and interpretability are identified. For un‐structured datasets, deep learning algorithms are explored and a critical view of the use of these approaches in radiation oncology is also provided. Conclusions We discuss the challenges in radiotherapy outcome prediction, and suggest to improve radiation outcome modeling by developing appropriate machine learning approaches where both accuracy and interpretability are taken into account.

show abstract

“…By using the model recommended by this study, it is possible to anticipate measured in vivo dose from TARGIT in the operating room. This work shows that the machine learning approach can be applied to in vivo dosimetry, as successfully done in other areas of radiotherapy workflow, including error detection and prevention, treatment dose planning, and verification . It also suggests that the results of in vivo dosimetry should be included among other categories of patient data in the electronic patient records in order to be available for automated data extraction with the aim of generating or improving prediction models for patient dose.…”

Section: Discussionmentioning

confidence: 86%

“…This tool would allow timely adoption of strategies to prevent excessive skin dose, such as placing a saline solution–soaked gauze as a spacer around the applicator, in order to increase source to skin distance. For this purpose, we use statistical and/or machine learning algorithms able to infer a hypothesis (the function/model), to predict the labels (skin dose) of out‐of‐sample observations . With the goal of achieving the best possible accuracy, the models are trained on data from in vivo skin dosimetry performed with an established technique on a large cohort of patients during more than 4 yr of TARGIT practice at our center.…”

Section: Introductionmentioning

confidence: 99%

“…For this purpose, we use statistical and/or machine learning algorithms able to infer a hypothesis (the function/model), to predict the labels (skin dose) of out-of-sample observations. 15,16 With the goal of achieving the best possible accuracy, the models are trained on data from in vivo skin dosimetry performed with an established technique on a large cohort of patients during more than 4 yr of TARGIT practice at our center.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Prediction of skin dose in low‐kV intraoperative radiotherapy using machine learning models trained on results of in vivo dosimetry

et al. 2019

View full text Add to dashboard Cite

Purpose The purpose of this study was to implement a machine learning model to predict skin dose from targeted intraoperative (TARGIT) treatment resulting in timely adoption of strategies to limit excessive skin dose. Methods A total of 283 patients affected by invasive breast carcinoma underwent TARGIT with a prescribed dose of 6 Gy at 1 cm, after lumpectomy. Radiochromic films were used to measure the dose to the skin for each patient. Univariate statistical analysis was performed to identify correlation of physical and patient variables with measured dose. After feature selection of predictors of in vivo skin dose, machine learning models stepwise linear regression (SLR), support vector regression (SVR), ensemble with bagging or boosting, and feed forward neural networks were trained on results of in vivo dosimetry to derive models to predict skin dose. Models were evaluated by tenfold cross validation and ranked according to root mean square error (RMSE) and adjusted correlation coefficient of true vs predicted values (adj‐R2). Results The predictors correlated with in vivo dosimetry were the distance of skin from source, depth‐dose in water at depth of the applicator in the breast, use of a replacement source, and irradiation time. The best performing model was SVR, which scored RMSE and adj‐R2, equal to 0.746 [95% confidence intervals (CI), 95% CI 0.737,0.756] and 0.481 (95% CI 0.468,0.494), respectively, on the tenfold cross validation. Conclusion The model trained on results of in vivo dosimetry can be used to predict skin dose during setup of patient for TARGIT and this allows for timely adoption of strategies to prevent of excessive skin dose.

show abstract

Machine Learning in Radiation Oncology: Opportunities, Requirements, and Needs

Cited by 93 publications

References 68 publications

Clinical implementation of magnetic resonance imaging guided adaptive radiotherapy for localized prostate cancer

Clinical implementation of magnetic resonance imaging guided adaptive radiotherapy for localized prostate cancer

Machine learning for radiation outcome modeling and prediction

Prediction of skin dose in low‐kV intraoperative radiotherapy using machine learning models trained on results of in vivo dosimetry

Contact Info

Product

Resources

About