Clinical evaluation of atlas and deep learning based automatic contouring for lung cancer

Lustberg, Tim; Soest, Johan van; Gooding, Mark J.; Peressutti, Devis; Aljabar, Paul; Stoep, J. van der; Elmpt, Wouter van; Dekker, André

doi:10.1016/j.radonc.2017.11.012

Cited by 271 publications

(243 citation statements)

References 26 publications

Supporting

Mentioning

238

Contrasting

Unclassified

Order By: Relevance

“…Full analysis of the results of this time assessment have previously be reported in Ref. [20], but are reported here as the real measure of clinical impact for which approaches such as DSC or the misclassification rate seek to act as a surrogate.…”

Section: Resultsmentioning

confidence: 99%

Comparative evaluation of autocontouring in clinical practice: A practical method using the Turing test

et al. 2018

Self Cite

View full text Add to dashboard Cite

A better correspondence with time saving was observed for the misclassification rate than the quantitative contour measures explored. From this, we conclude that the inability to accurately judge the source of a contour indicates a reduced need for editing and therefore a greater time saving overall. Hence, task-based assessments of contouring performance may be considered as an additional way of evaluating the clinical utility of autosegmentation methods.

show abstract

Section: Resultsmentioning

confidence: 99%

Comparative evaluation of autocontouring in clinical practice: A practical method using the Turing test

et al. 2018

Self Cite

View full text Add to dashboard Cite

show abstract

“…Possible solutions include resampling the images to lower resolution at the sacrifice of high‐resolution details and edge information; building a shallow network, which will reduce the power of the CNN model; extract smaller patches from the input images for network training, which may also reduce the model accuracy due to the loss of image information. A few studies have been proposed to combine 2D/3D deep learning networks with traditional segmentation algorithms such as 3D majority voting, random walk or atlas‐based methods or use collaborative networks to overcome these issues and improve the segmentation accuracy. However, traditional segmentation algorithms may suffer from robustness issues when combined with deep learning methods as they may worsen the outputs of the neural networks on images with suboptimal image quality or reduced contrast, where the hypothesis for such algorithms is generally no longer valid.…”

Section: Introductionmentioning

confidence: 99%

Deep convolutional neural network for segmentation of thoracic organs‐at‐risk using cropped 3D images

et al. 2019

View full text Add to dashboard Cite

Purpose Automatic segmentation of organs‐at‐risk (OARs) is a key step in radiation treatment planning to reduce human efforts and bias. Deep convolutional neural networks (DCNN) have shown great success in many medical image segmentation applications but there are still challenges in dealing with large 3D images for optimal results. The purpose of this study is to develop a novel DCNN method for thoracic OARs segmentation using cropped 3D images. Methods To segment the five organs (left and right lungs, heart, esophagus and spinal cord) from the thoracic CT scans, preprocessing to unify the voxel spacing and intensity was first performed, a 3D U‐Net was then trained on resampled thoracic images to localize each organ, then the original images were cropped to only contain one organ and served as the input to each individual organ segmentation network. The segmentation maps were then merged to get the final results. The network structures were optimized for each step, as well as the training and testing strategies. A novel testing augmentation with multiple iterations of image cropping was used. The networks were trained on 36 thoracic CT scans with expert annotations provided by the organizers of the 2017 AAPM Thoracic Auto‐segmentation Challenge and tested on the challenge testing dataset as well as a private dataset. Results The proposed method earned second place in the live phase of the challenge and first place in the subsequent ongoing phase using a newly developed testing augmentation approach. It showed superior‐than‐human performance on average in terms of Dice scores (spinal cord: 0.893 ± 0.044, right lung: 0.972 ± 0.021, left lung: 0.979 ± 0.008, heart: 0.925 ± 0.015, esophagus: 0.726 ± 0.094), mean surface distance (spinal cord: 0.662 ± 0.248 mm, right lung: 0.933 ± 0.574 mm, left lung: 0.586 ± 0.285 mm, heart: 2.297 ± 0.492 mm, esophagus: 2.341 ± 2.380 mm) and 95% Hausdorff distance (spinal cord: 1.893 ± 0.627 mm, right lung: 3.958 ± 2.845 mm, left lung: 2.103 ± 0.938 mm, heart: 6.570 ± 1.501 mm, esophagus: 8.714 ± 10.588 mm). It also achieved good performance in the private dataset and reduced the editing time to 7.5 min per patient following automatic segmentation. Conclusions The proposed DCNN method demonstrated good performance in automatic OAR segmentation from thoracic CT scans. It has the potential for eventual clinical adoption of deep learning in radiation treatment planning due to improved accuracy and reduced cost for OAR segmentation.

show abstract

“…However, it should be noticed that most of them rely on the accuracy of DIR in the contour generation (e.g., via atlases). More recent machine learning algorithms do not, but they lack patient‐specific validation, and their generalization capabilities beyond the training dataset are uncertain . Therefore the use of automatic contouring to validate DIR on a patient‐specific basis is quite questionable and is often not able to match the accuracy of the expert clinician, which remains the universally acknowledged gold standard .…”

Section: The Geometric Accuracy Paradigmmentioning

confidence: 99%

“…More recent machine learning algorithms do not, but they lack patientspecific validation, and their generalization capabilities beyond the training dataset are uncertain. 93,94 Therefore the use of automatic contouring to validate DIR on a patient-specific basis is quite questionable and is often not able to match the accuracy of the expert clinician, which remains the universally acknowledged gold standard. 92 Moreover, it would still suffer of the same limitations of the above-mentioned contour-based metrics.…”

Section: B Automatic Strategies: Image-basedmentioning

confidence: 99%

“Patient‐specific validation of deformable image registration in radiation therapy: Overview and caveats”

et al. 2018

View full text Add to dashboard Cite

Over the last few decades, deformable image registration (DIR) has gained popularity in image-guided radiation therapy for a number of applications, such as contour propagation, dose warping, and accumulation. Although this raises promising perspectives for the improvement of treatment outcomes and quality of radiotherapy clinical practice, the variety of proposed DIR algorithms, combined with the lack of an effective quantitative quality control metric of the registration, is slowing the transfer of DIR into the clinical routine. Recently, a task group (AAPM TG132) report was published outlining the essential aspects of DIR for image guidance in radiotherapy. However, an accurate and efficient patient-specific validation is not yet defined, and appropriate metrics should be identified to achieve the definition of both geometric and dosimetric accuracy. In this respect, the use of a dense set of anatomical landmarks, along with additional evaluations on contours or deformation field analysis, are likely to drive patient-specific DIR validation in clinical image-guided radiotherapy applications to account for geometric inaccuracies. Automatic and efficient strategies able to provide spatial information of DIR uncertainties and to evaluate monomodal and multimodal image registration, as well as to describe homogenous and un-contrasted regions are believed to represent the future direction in DIR validation. But especially in the case of DIR applications for dose mapping and accumulation, the need of accurate patient-specific validation is not only limited to the evaluation of geometric accuracy. In fact, the need to account for dosimetric inaccuracies due to DIR represents another important area in the field of adaptive treatments. Different approaches are currently being investigated to quantify the effect of DIR error on dose analysis, mainly relying on clinically relevant dose metrics, or on the study of deformation field properties for a voxel-by-voxel evaluation. However, novel research is required for the definition of dedicated and personalized measures capable to relate the geometric and dosimetric inaccuracies, thus bearing useful information for a safe use of DIR by clinical end users. In this paper we provide insights on DIR results evaluation on a patient-specific basis, facing the issues of both geometric and dosimetric paradigms. Challenges on DIR validation are overviewed and discussed, in order to push preliminary clinical guidelines forward on this fundamental topic and boost the implementation of more robust and reliable patient-specific evaluation metrics.

show abstract

Clinical evaluation of atlas and deep learning based automatic contouring for lung cancer

Abstract: User adjustment of software generated contours is a viable strategy to reduce contouring time of OARs for lung radiotherapy while conforming to local clinical standards. In addition, deep learning contouring shows promising results compared to existing solutions.

Cited by 271 publications

References 26 publications

Comparative evaluation of autocontouring in clinical practice: A practical method using the Turing test

Comparative evaluation of autocontouring in clinical practice: A practical method using the Turing test

Deep convolutional neural network for segmentation of thoracic organs‐at‐risk using cropped 3D images

“Patient‐specific validation of deformable image registration in radiation therapy: Overview and caveats”

Contact Info

Product

Resources

About