Skip to Main Content
 

Global Search Box

 
 
 
 

ETD Abstract Container

Abstract Header

FDG-PET/MR for Cervical Cancer Staging and Radiation Therapy Planning: A Novel, Deep Learning-based Approach

Abstract Details

2020, Doctor of Philosophy, Case Western Reserve University, Biomedical Engineering.
Multi-modality imaging is on the forefront of precision medicine, especially in the field of oncology, where radiological data enables accurate staging, prognostication, and precise anatomical localization for radiation therapy planning. In cervical cancer for example, 18F-labelled 2-fluoro-2-deoxy-D-glucose (FDG) positron emission tomography (PET) allows precise quantification of the local disease extent and distant metastatic burden. Moreover, magnetic resonance (MR) images empower a careful cancerous tissue targeting with radiation therapy by yielding a superior soft tissue contrast. As for computed tomography (CT), it supports FDG-PET attenuation correction and radiation planning by providing electron density information. However, such comprehensive and extensive investigations are generally expensive, inconvenient for the patient, and labor-intensive for the medical professionals especially for the radiation oncologist who needs to perform manual contouring of the targeted tumor and the adjacent normal tissues to be avoided or organs at risk (OARs). Nonetheless, the accuracy of such approach is affected by the inherent multi-image registration errors, and a workflow based on a single imaging session is highly desired. In this dissertation, we designed a novel approach for FDG-PET/MR based cervical cancer staging and radiation planning, by tailoring the state of art deep learning algorithms in order to accelerate and automate the workflow. First, we took advantage of image to image translation capacity with deep learning to perform cross-modality image synthesis by generating synthetic CT (sCT) from MR images, thus precluding the need for CT acquisition. To address time-expenses and patients discomfort without jeopardizing the treatment effectiveness, we carefully built our sCT generation models on two MR sequences only. Finally, we developed a novel approach for automatic contouring that leverage general anatomical topography knowledge with the deep learning semantic segmentation capacity. The dissertation explores a new tactic to address data availability for deep learning in medicine, wherein compacted architecture of deep convolutional networks can outperform the usual architecture when appropriately manipulated by the human user. From a clinical perspective, the presented workflow is expected to expedite the cervical cancer treatment, and can be applied in conventional and adaptive radiation therapy settings.
Raymond Muzic, Jr, Ph.D. (Advisor)
David Wilson, Ph.D. (Committee Chair)
Robert Brown, Ph.D. (Committee Member)
Xin Yu, Sc.D. (Committee Member)
Bryan Traughber, M.D. (Committee Member)
129 p.

Recommended Citations

Citations

  • Baydoun, A. (2020). FDG-PET/MR for Cervical Cancer Staging and Radiation Therapy Planning: A Novel, Deep Learning-based Approach [Doctoral dissertation, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1594844980840027

    APA Style (7th edition)

  • Baydoun, Atallah. FDG-PET/MR for Cervical Cancer Staging and Radiation Therapy Planning: A Novel, Deep Learning-based Approach. 2020. Case Western Reserve University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1594844980840027.

    MLA Style (8th edition)

  • Baydoun, Atallah. "FDG-PET/MR for Cervical Cancer Staging and Radiation Therapy Planning: A Novel, Deep Learning-based Approach." Doctoral dissertation, Case Western Reserve University, 2020. http://rave.ohiolink.edu/etdc/view?acc_num=case1594844980840027

    Chicago Manual of Style (17th edition)