Radiation oncology has for several decades been a field with constantly evolving technological developments. Technology has contributed to form our evidence-based scientific discipline determining the most favorable strategies for delivering radiotherapy with optimal radiation doses at the right time and place to achieve the optimal outcome [1]. Technological developments in medical image acquisition and analysis have increasingly provided faster and more detailed anatomical imaging and are today central for contouring of both targets and organs at risk (OARs), treatment planning, response prediction and evaluation, and quality assurance. On the other side, errors in image acquisition and quantification impact directly on the accuracy of radiotherapy delivery. Two papers exploiting technological advancements in imaging to develop new and more automated strategies for OAR and metastatic lymph node contouring have recently been published in our journal [2,3]. Computed tomography (CT) scanning has been pivotal in the development of radiotherapy planning. CT, now often acquired daily during the course of treatment, provides the geometric fidelity required to assess the position of the tumor, surrounding tissues and OARs. However, with the increasing availability and integration of magnetic resonance imaging (MRI) into the radiotherapy planning process, additional opportunities are emerging. MRI offers a superior soft-tissue contrast compared to CT, and also provides possibilities for a multitude of different acquisition protocols enabling both detailed anatomical imaging and assessment of a range of functional tissue properties using the more advanced protocols such as diffusion-weighted MRI and dynamic contrast-based MRI. These functional sequences can together with post-processing tools provide quantitative measures of radiobiological tissue characteristics, which can be exploited to deliver more tailored radiotherapy to each patient and also with the possibility for adjustments during the course of treatment [4]. The use of MRI may therefore result in a more optimized treatment where the tumor response is increased and normal tissue damage is decreased. With the recent developments of novel MR-guided radiotherapy systems, including the integration of MRI scanners and linear accelerators, MRI is now becoming a reality also for daily monitoring of geometric accuracy, dose accumulation and of radiotherapy response measures [5,6]. Although MRI has been a more resource and time-demanding acquisition than CT, new technological developments with for instance parallel acquisition are now providing faster, high-resolution, 4-dimensional acquisitions providing both anatomical and functional information during radiotherapy delivery [7,8]. This opens new avenues for quantitative assessment of longitudinal changes during the course of radiotherapy. A key challenge for daily quantitative imaging is contouring of the target and OARs. Currently, expert clinicians perform the contouring manually. Such a process is labor-intensive and also associated with considerable variations between experts. Contouring accuracy is regarded a particularly important task in radiation oncology, as suboptimal tumor coverage and poor quality radiotherapy plans are major factors for disease relapse and inferior survival [9]. Automation of the contouring process has the potential to substantially decrease the workload while possibly increasing contour consistency [10]. The need for automated contouring of the target and OARs has been a major reason for why artificial intelligence (AI) has become attractive for our discipline [11]. AI and machine learning are terms used to describe computerized approaches to identify complex mathematical relationships within data. While AI is not a new concept, recent advances in computing power, algorithms, data collection and data sharing have enabled an explosion in the capabilities and utilization of AI. This is also facilitated by an increase in parallel computing capabilities through graphics processing unit (GPU) architectures and other frameworks such as cloud-based computing. Krizhevsky et al. presented the breakthrough study in 2012 using a convolutional neural network (CNN) model, AlexNet, to reduce the error rate for object (i.e. target and OAR) recognition [12]. This model showed impressive results and became important for further developments of organ segmentation in radiotherapy. Later, Tong et al. used CNN models to perform automatic multi-organ segmentation in patients with head and neck cancer [13]. However, for clinical use in radiotherapy planning, automated target and OAR segmentation needs to be robust and accurate. Recent studies investigated the use of different networks to reach maximum accuracy for automatic segmentation [14]. Efficient translation of the methods to other centers has to be guaranteed. In this issue, Brunenberg et al. performed an independent validation of a deep learning-based CT contouring method for OARs in the head-and-neck region [2]. The study demonstrated that AI-based automatic contouring which had been trained in one institution could safely and efficiently be transferred to another institution for subsequent clinical use. Such independent validation is of crucial importance to ensure freedom from dependencies on institutional image acquisition settings. Further in this issue, Gurney-Champion et al. combined 3D CNN models with quantitative information from diffusion-weighted MR images to achieve automatic contouring of metastatic lymph nodes in patients with head and neck cancer [3]. This study aimed at developing a highly reproducible method for lymph node segmentation in order to objectively analyze sequential information from quantitative information assessed during each fraction of radiotherapy delivery. With this visionary approach, the study paves
[1]
Wei Lu,et al.
Quantification of accumulated dose and associated anatomical changes of esophagus using weekly Magnetic Resonance Imaging acquired during radiotherapy of locally advanced lung cancer
,
2020,
Physics and imaging in radiation oncology.
[2]
B Stemkens,et al.
Nuts and bolts of 4D-MRI for radiotherapy
,
2018,
Physics in medicine and biology.
[3]
Brent van der Heyden,et al.
Evaluation of measures for assessing time-saving of automatic organ-at-risk segmentation in radiotherapy
,
2019,
Physics and imaging in radiation oncology.
[4]
Nuo Tong,et al.
Fully automatic multi‐organ segmentation for head and neck cancer radiotherapy using shape representation model constrained fully convolutional neural networks
,
2018,
Medical physics.
[5]
K. Harrington,et al.
A convolutional neural network for contouring metastatic lymph nodes on diffusion-weighted magnetic resonance images for assessment of radiotherapy response
,
2020,
Physics and imaging in radiation oncology.
[6]
Geoffrey E. Hinton,et al.
ImageNet classification with deep convolutional neural networks
,
2012,
Commun. ACM.
[7]
Harini Veeraraghavan,et al.
Deep learning-based auto-segmentation of targets and organs-at-risk for magnetic resonance imaging only planning of prostate radiotherapy
,
2019,
Physics and imaging in radiation oncology.
[8]
Tom Bruijnen,et al.
A dual-purpose MRI acquisition to combine 4D-MRI and dynamic contrast-enhanced imaging for abdominal radiotherapy planning
,
2019,
Physics in medicine and biology.
[9]
Yeon Soo Yeom,et al.
Automatic segmentation of cardiac structures for breast cancer radiotherapy
,
2019,
Physics and imaging in radiation oncology.
[10]
L. Holloway,et al.
Uncertainties in volume delineation in radiation oncology: A systematic review and recommendations for future studies.
,
2016,
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology.
[11]
Mechthild Krause,et al.
Radiation oncology in the era of precision medicine
,
2016,
Nature Reviews Cancer.
[12]
G. G. Sikkes,et al.
Target coverage and dose criteria based evaluation of the first clinical 1.5T MR-linac SBRT treatments of lymph node oligometastases compared with conventional CBCT-linac treatment.
,
2020,
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology.
[13]
Gilmer Valdes,et al.
Artificial intelligence in radiation oncology: A specialty-wide disruptive transformation?
,
2018,
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology.
[14]
Daniela Thorwarth,et al.
Quantitative imaging for radiotherapy purposes
,
2020,
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology.
[15]
Marko Pesola,et al.
Validation of automated magnetic resonance image segmentation for radiation therapy planning in prostate cancer
,
2020,
Physics and imaging in radiation oncology.
[16]
Charlotte L. Brouwer,et al.
External validation of deep learning-based contouring of head and neck organs at risk
,
2020,
Physics and imaging in radiation oncology.