(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose:
In this study, we investigate whether an interactive contouring approach can leverage context information using deep learning to assist clinicians in manual contouring tasks.
Methods:
A deep learning segmentation model, using a U-Net architecture, was trained using two alternative approaches. The models were trained either including contours from a single organ or from a variety of organs. Contextual information was provided to the model, using the prior contoured slice as an input, in addition to the slice to be contoured. The AAPM 2017 Thoracic contouring challenge dataset was used. Each case contains five OARs: heart, left and right lung, oesophagus and spinal cord. 12082 contoured organ slices were used for slice-by-slice training. Results were evaluated on 4647 slices using DICE similarity coefficient. Both models, were evaluated on all OARs, regardless of the training set used.
Results:
The DICE for heart segmentation was 0.91 for the single-organ model that was trained using a heart-only training set. However, the mean DICE for the other 4 OARs was 0.025. For a multi-organ model, the DICE on heart segmentation was 0.92 with a mean DICE of 0.76 for the other OARs.
Conclusion:
The single-organ model can only contour the organ it has been trained on. But fails to contour organs outside the training set despite being provided context information. A model trained on various organs learns to predict different organs based on the context between CT and corresponding contour. This study has demonstrated that user provided context can be incorporated into deep learning contouring to facilitate semi-automatic segmentation of CT imaging. An appropriate training set must be selected to ensure that the approach generalises to use prior context rather than learning organ-specific segmentation. Such an approach may enable faster de-novo contouring in clinical practice.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by the European Unions Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 766,276. This paper reflects only the authors views and the European Commission Research Executive Agency is not responsible for any use that may be made of the contained information
Keywords
Computer Software, Image Processing, Segmentation
Taxonomy
IM/TH- Image Segmentation Techniques: Machine Learning
Contact Email
