Insights into deep learning methods with application to cancer imaging

Huttinga, N.R.F. (2017) Insights into deep learning methods with application to cancer imaging.

[img]
Preview
PDF
5MB
Abstract:In recent years, deep learning methods have received a massively increasing amount of attention due to their almost human-level performance on high-level vision tasks like semantic segmentation, object classification and automatic image captioning. These methods have also shown to be competitive with current state-of-the-art methods for low-level vision tasks like denoising and inpainting. Despite their impressive performance, deep learning methods still lack a unified theoretical foundation and understanding. Low-level vision tasks can, on the other hand, also be addressed by methods that do have a very strong theoretical foundation. In this work we explore similarities between the two approaches for low-level vision task to gain more insight into deep learning methods in general. We explain the performance of convolutional neural networks (CNNs), and show that denoising auto-encoders effectively solve a bi-level optimization problem. On top of this, we apply the insights into CNNs to images of circulating tumor cells (CTCs). More specifically, a CNN is trained to perform classification of CTCs based on high-throughput fluorescence microscopy images obtained from the medical cell biophysics group (MCBP) at the University of Twente.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:31 mathematics, 44 medicine, 54 computer science
Programme:Applied Mathematics MSc (60348)
Link to this item:http://purl.utwente.nl/essays/72366
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page