Year: 2021
Type: dissertation
Abstract: Multiparametric MRI based assessment of renal masses in the kidney has the potential to improve tumour classification accuracy and as a result improve treatment outcomes for renal cancer patients. Despite the increased use of MRI for renal mass assessment, the use of deep learning techniques to detect and classify renal masses using this modality remains unexplored. I propose a fully automated computer aided detection algorithm that identifies spatially separated renal masses in abdominal contrast-enhanced nephrographic phase MRI volumes. A cascaded series of U-Net models is used, with the first step isolating kidney boundaries with a Dice score of 91.20%. These boundaries are used to identify instances of renal masses with a mass-wise precision of 83.3% and a recall of 86.2%. The proposed algorithm can serve as a localizer for future work that incorporates multiparametric MRI data to classify tumours based on pathology.
Author: Agarwal, Anush
Institution: University of Guelph
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7026910789