Multi-label Classification to Hierarchical Taxonomies for Lung
Principal Investigator
Name
cai chen
Degrees
M.D.
Institution
Qingdao University of Science and Technology
Position Title
Master
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCOI-704
Initial CDAS Request Approval
Dec 22, 2020
Title
Multi-label Classification to Hierarchical Taxonomies for Lung
Summary
In addition to lung-related diseases, lung diseases can also cause a variety of complications, such as bronchitis, heart disease, and lymphatic system diseases. We are designing a system that takes an approach incorporating top-down knowledge to model the conditional dependence of child labels upon their parents that allow the model to focus on discriminating between siblings rather than across all disease patterns. And as popular now, we attempt to train classifiers to predict conditional probabilities at each node and particularly focus on multi-label classification.
Aims
1. We train a classifier to predict conditional probabilities.
2. We use a DenseNet-121 model as a backbone trained jointly by all the tasks.
3. We try to allow a model to predict high confidence for non-leaf label predictions but lower confidence for leaf label predictions.
Collaborators
Qingdao University of Science and Technology