Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Lanjun Wang
Degrees
Ph.D
Institution
Tianjin University
Position Title
Researcher
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-956
Initial CDAS Request Approval
Sep 6, 2022
Title
Data Intrinsic Bias Detection in Medical Image Dataset
Summary
Machine learning has a significant role in several high-impact applications in the medical imaging domains, such as computer-aided diagnosis, image segmentation, registration and fusion, image-guided therapy, image annotation, and image database retrieval. Although medical imaging analysis has witnessed impressive progress in recent years thanks to the development of large-scale labeled datasets, it may result in major discrimination if not dealt with proper care on biases within such data. The bias in the data causes generalizability degradation of the developed models, which has been identified as one of the major limitations in deep learning applications in healthcare. In this study, we mainly focus on investigating the intrinsic biases in medical image datasets based on machine learning and data mining techniques. We first carefully define the data intrinsic bias attribute, and then propose a novel bias identification framework for medical image datasets. The framework contains two major components, KlotskiNet and Bias Discriminant Direction Analysis (BD^2A), where KlostkiNet is to build the mapping which makes backgrounds to distinguish positive and negative samples and BD^2A provides a theoretical solution for determining bias attributes. More datasets are required in the experiments.
Aims

* Identify intrinsic bias in the medical image
* Identify the root cause of bias in machine learning models
* Evaluate intrinsic bias of dataset and build up debiased model

Collaborators

Qian Zhao, School of Basic Medical Science, Tianjin Medical University