Medical Image Analysis
Medical Image Analysis
Medical imaging includes a number of different techniques, ranging from computerized tomography and magnetic resonance imaging to multi- and hyperspectal images. Such imaging techniques create visual representation of a patients interior, which physicians can make use of to perform clinical analysis. Medical images can be challenging and time-consuming to analyze, which is why there has been many studies on designing automatic systems that can aid physicians during analysis. The UiT Machine Learning is at the very front of development of such systems, with a particular focus on utilizing recent deep learning methodology.
Lung cancer is the most frequent cancer type and the leading cause of cancer-related death in the world. Early and accurate staging of the disease is important when deciding which treatment should be given. The UiT Machine Learning Group is currently heading a project focused developing innovative classification and segmentation methods for lung cancer by taking advantage of biological information from magnetic resonance images. The foundation of the project is deep learning methodology and, in particular, convolutional neural networks.
Another leading cause of cancer-related death worldwide is colorectal cancer. Utilizing novel deep learning methodology, our group has developed a decision support systems tasked with aiding physicians during the early detection and prevention of colorectal cancer, with a particular focus on creating interpretable and transparent models.
Yet another common cause of cancer-death is skin cancer, where the incidence trend for the past 30 years has been increasing. Early detection and prevention is crucial for increasing survival rates, and one promising direction for achieving these goals is through the analysis of multi- and hyperspectal images. As a result of the high dimensionality of such images, they are challenging to process and analyze. The UiT Machine Learning group is comining deep learning with advanced statistical methods in order to overcome the challenges assoicated with multi- and hyperspectal images.
- Kristoffer Knutsen Wickstrøm, Michael Kampffmeyer and Robert Jenssen
- Samuel Kuttner, Kristoffer Knutsen Wickstrøm, Gustav Kalda, S Esmaeil Dorraji, Montserrat Martin-Armas, Ana Oteiza, Robert Jenssen, Kristin Fenton, Rune Sundset and Jan Axelsson