Deep learning-based organ-at-risk segmentation in head-and-neck radiotherapy

Abstract

This doctoral thesis is the product of scientific research conducted from early 2018 to early 2021. It covers work investigating the potential of deep learning-based segmentation for organ-at-risk segmentation in head and neck radiotherapy. Deep learning-based segmentation of organs-at-risk in head and neck radiotherapy can be used to generate treatment plans that are as safe as treatment plans based on manual segmentation. However, in our study, deep learning-based segmentation did not yet perform well enough to be very specific during quality assurance of manual segmentation. There are several strategies that can be applied to improve the performance and reliability of deep learning-based segmentation, but there seems to be an upper limit on the similarity coefficient that can be achieved. This may be caused by the sub-optimal quality of the manual segmentations used to train deep learning models. Contrarily, sub-optimal quality of the imaging data can make the model more robust. The imperfect performance of deep learning-based segmentation may be one of the reasons that it is not yet standard practice in radiotherapy clinics around the world. Probability maps may be a way to increase user-confidence and facilitate adoption of these methods in the clinic.

Related Papers

• → Overview of deep learning in medical imaging(2017)1,052 cited
• → A Survey on Deep Learning Architectures and Frameworks for Cancer Detection in Medical Images Analysis(2020)4 cited
• → Deep fake Detection Through Deep Learning(2023)4 cited
• Why & When Deep Learning Works: Looking Inside Deep Learnings.(2017)
• → Why & When Deep Learning Works: Looking Inside Deep Learnings(2017)