Active Learning for Brain Tumor Segmentation

Shen, Maohao

Permalink

https://hdl.handle.net/2142/110319 Copy

Description

Title

Active Learning for Brain Tumor Segmentation

Author(s)

Shen, Maohao

Contributor(s)

Koyejo, Oluwasanmi

Issue Date

2021-05

Keyword(s)

• Active Learning
• Segmentation
• Deep Learning
• Uncertainty
• Approximation Algorithm

Abstract

Over the last decade, deep learning has achieved tremendous progress in many fields. However, the performance of deep learning models relies on vast datasets for training, and access to labeled training data is among the most pressing roadblocks in a lot of real-world applications. Therefore, saving labeling costs while still efficiently training a deep learning model becomes a meaningful research problem. Active Learning (AL) is an established framework designed to mitigate the problem of scarce labeled data. In this work, we study Active Learning from biomedical imaging application perspectives. For the biomedical image segmentation problem, the difficulty of obtaining sufficient labeled data can be a bottleneck. To this end, we design a novel active learning framework specially adapted to brain tumor segmentation. Our approach includes a novel labeling cost designed to capture radiologists' practical labeling costs. This is combined with two acquisition functions to incorporate uncertainty and representation information, ensuring that the active learning selects informative and diverse data. The resulting procedure is a constrained combinatorial optimization problem. We propose an efficient algorithm for this task and demonstrate the proposed method's advantages for segmenting brain MRI data.

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/110319

Owning Collections