10. 1. many types 2. first-timer 3. label effort
Construct cell Learn from
size distribution previous types ?
11. Training Image
Cell
Non-
cell
Training Samples
User
Size Distribution
GATLAB
label effort
random
interactive
Previous types
Select most important
samples for user to label.
12. Training Image
Cell
Non-
cell
Training Samples
User
Size Distribution
Detection Confidence
GATLAB
Previous types
13.
14. White Blood Cells HT29 Cancer Natural Killer T Drosophila Red Blood Cells
15. AdaBoost uses Adaptive Boosting
TaskTrAdaBoost learns from previous cell types
GlobalTrAdaBoost obtains cell size distribution
GATLAB selects most important samples
Freund and Schapire (2000)
Yao and Doretto (2010)
Nguyen et al. (2011)
25. An accurate cell detection algorithm.
Require minimal training effort.
Help biologists to study various cell types.
26. N. Nguyen, E. Norris, M. Clemens, M. Shin. “Rapidly Adaptive Cell Detection.”
Machine Vision and Applications (MVA), Special Issue: Machine Learning in
Medical Imaging [in review].
N. Nguyen and M. Shin. “Active Transfer Boosting to Reduce Training Effort in
Multi-class Data classification." IEEE International Conference on Computer
Vision and Pattern Recognition (CVPR), Providence, Rhode Island, June 18-20, 2012
[in review].
N. Nguyen, E. Norris, M. Clemens, M. Shin. “Rapidly Adaptive Cell Detection using
Transfer Learning with a Global Parameter.” The Second International
Workshop on Machine Learning in Medical Imaging (MLMI), Toronto, Canada.
September 18-22, 2011.
N. Nguyen, S. Keller, E. Norris, T. Huynh, M. Clemens, M. Shin. “Tracking Colliding
Cells in vivo Microscopy Video.” IEEE Transactions on Biomedical Engineering
(TBE), 58(8):2391-2400, August 2011.
N. Nguyen, S. Keller, T. Huynh, M. Shin. “Tracking Colliding Cells”. IEEE Workshop
on Applications of Computer Vision (WACV), Snowbird, UT December 07-09,
2009.
27. Min Shin, PhD Mark Clemens, PhD Eric Norris, MS Toan Huynh, MD Steve Keller, MS
Notes de l'éditeur
A special type of white blood cells, call natural killer t-cells, has a potential of killing cancer tumor.
10 training samples, which is only 10% of the training effort as
Our previous research has solved the first 2 of these challenges.
Elaborate much more in this one.
Elaborate much more in this one.
Need to have all four methods. 1 figure that said it all. Zoom in on the 1 to 10 number of training samples.
Need to have all four methods. 1 figure that said it all. Zoom in on the 1 to 10 number of training samples.