Celebrating the defence
Image:
Photo: Changkyu Choi

Celebrating the defence

Excellent PhD defense in the Visual Intelligence research centre

Stine Hansen with the defended her PhD thesis “Leveraging Supervoxels for Medical Image Volume Segmentation With Limited Supervision” on Dec. 16th at UiT The Arctic University of Norway.

Excellent PhD defense in the Visual Intelligence research centre

The focus of Stine Hansens thesis is on the development of new ML algorithms for medical image segmentation with limited supervision. In particular, completely unsupervised solutions to lung tumor segmentation and solutions that only require a few labeled samples to perform organ segmentation (few-shot learning) are proposed. To exploit the data-specific opportunities of medical images, automatically generated supervoxels are exploited in various ways to solve key challenges related to the task.

Dr. Stine Hansen (Center) with family, committee and supervisor. Photo: Changkyu Choi

Evaluation Committee

  • Professor Alexandros Iosifidis, Department of Electrical and Computer Engineering, Aarhus University, Danmark (1. Opponent)
  • Professor Kjersti Engan, Institutt for data og teknologi, Universitetet i Stavanger (2. Opponent)
  • Associate professor Shujian Yu, IFT, UiT (internal member)

Supervisors

  • Professor Robert Jenssen, IFT, UiT (main supervisor)
  • Associate professor Stian Normann Anfinsen, IFT, UiT (co-supervisor)

Summary of Stine Hansen's thesis work:

The majority of existing methods for machine learning-based medical image segmentation are supervised models that require large amounts of fully annotated images. These types of datasets are typically not available in the medical domain and are difficult and expensive to generate. A wide-spread use of machine learning based models for medical image segmentation therefore requires the development of data-efficient algorithms that only require limited supervision. To address these challenges, this thesis presents new machine learning methodology for unsupervised lung tumor segmentation and few-shot learning based organ segmentation. When working in the limited supervision paradigm, exploiting the available information in the data is key. The methodology developed in this thesis leverages automatically generated supervoxels in various ways to exploit the structural information in the images. The work on unsupervised tumor segmentation explores the opportunity of performing clustering on a population-level in order to provide the algorithm with as much information as possible. To facilitate this population-level across-patient clustering, supervoxel representations are exploited to reduce the number of samples, and thereby the computational cost. In the work on few-shot learning-based organ segmentation, supervoxels are used to generate pseudo-labels for self-supervised training. Further, to obtain a model that is robust to the typically large and inhomogeneous background class, a novel anomaly detection-inspired classifier is proposed to ease the modelling of the background. To encourage the resulting segmentation maps to respect edges defined in the input space, a supervoxel-informed feature refinement module is proposed to refine the embedded feature vectors during inference. Finally, to improve trustworthiness, an architecture-agnostic mechanism to estimate model uncertainty in few-shot segmentation is developed. Results demonstrate that supervoxels are versatile tools for leveraging structural information in medical data when training segmentation models with limited supervision.

Full Thesis Here

Latest news

Report from the VIGS gathering in Tromsø

December 1, 2022

7-8 November VIGS organized a successful gathering for early career researchers in Tromsø. The program was filled with both a social and technical program and gathered around 20 researchers from the research partners in Visual Intelligence.

Visual Intelligence visits by Lars Kai Hansen and Irina Voiculescu

November 16, 2022

We had the great pleasure of hosting Lars Kai Hansen, Technical University of Denmark, and Irina Voiculescu, University of Oxford, at the Visual Intelligence centre in Tromsø.

Successful PhD defense in the Visual Intelligence research centre

November 14, 2022

Kristoffer Wickstrøm defended his PhD thesis “Advancing deep learning with emphasis on data-driven healthcare” on Oct 28 at UiT The Arctic University of Norway.

Visual Intelligence paper accepted to NeurIPS

September 23, 2022

PhD student Srishti Gautam in Visual Intelligence and her collaborators from UiT The Arctic University of Norway and Technical University of Berlin got their paper "ProtoVAE: A Trustworthy Self-Explainable Prototypical Variational Model" accepted to NeurIPS 2022!

Great success for Visual Intelligence at Arendalsuka

August 18, 2022

It was absolutely packed with people when Visual Intelligence, NORA, and CAIR/UIA invited a strong panel to debate the future of healthcare with AI at Arendalsuka 2022!

Visual Intelligence guest researcher Ane Blazquez Garcia defends her PhD

July 30, 2022

Ane Blazquez Garcia, University of the Basque Country and the Ikerlan research institute, visited Visual Intelligence in the fall of 2021. Partially based on work conducted during this visit, she successfully defended her PhD thesis on July 27th 2022. Congratulations!

Visual Intelligence keynote at the 2022 Norwegian AI Society Conference

June 9, 2022

Director Robert Jenssen gave the keynote "Visual Intelligence advances deep learning research towards innovations" at the 2022 Norwegian AI Society Conference (NAIS) in Oslo.

Pitch-day for ideas for master’s projects for the computer science and machine learning students at UiT

April 19, 2022

We invited companies, the public sector, and other organizations to present their ideas for master’s projects and to connect with students having an interest and background in computer science and machine learning.

Talk about Ethics in AI at the Aim North 2022 conference

April 6, 2022

Director Robert Jenssen gave a talk about explainable AI connected to the overarching theme of the Aim North conference, namely Ethics in AI. The talk outlined e.g. VI's recent results on detecting artifacts in medical images by our "Prototypical Relevance Propagation" method.