MCS Thesis Examination: Prabhakara Subramanya Jois

Abstract

    Modern surgery today has greatly improved healthcare due to technological advancements in medical imaging. It has fostered a culture of innovation that has progressed with continuous and incremental changes towards curing patients’ ailments. With evidence-based assessments gaining prominence in modern surgery, Radiomics has become crucial to resolving problems through the integration of the best scientific data with the correct clinical expertise. As a quantitative approach to medical imaging, Radiomics uses mathematical analysis to improve the data made available to clinicians, which greatly influences their decision-making ability. In this thesis, we focus on two applications: pelvic bone segmentation from CT data for designing patient-specific customizable pessaries; and quantitative assessment of breast morphology, for reconstructive breast surgeries.

    For pelvic bone segmentation, we investigate several encoder-decoder network configurations trained on limited data and use histogram based features from Radiomics to take a data-centric view towards the problem and boost the model performance on completely unseen data through histogram specification. Then we evaluate the performance on two publicly available CT datasets.

    For assessment of breast morphology, we propose a novel metric for quantifying the overall dissimilarity between two breast mounds, called VIMA, by using shape and size based features from iso-contours. The methodology was experimented on 3D scans of artificial breasts and found to be highly useful in an intra-operative setting for aiding surgeons during aesthetic breast surgeries.

Examining Committee

• Dr. Yiming Xiao (Chair) 
• Dr. Thomas Fevens (Supervisor)
  
• Dr. Marta Kersten (Examiner)
• Dr. Yiming Xiao (Examiner)