Blind Image Recovery [https://arxiv.org/pdf/1811.08484.pdf]
How can we leverage GAN priors to solve challenging inverse imaging problems without ground truth?
3D conv-nets for lung nodule detection [SPIE-MI 2016 paper]
Lung nodule detection using 3D convolutional filters, which can take a volume of the CT scan and predict the probability of a nodule being present. We proposed a multi-scale approach to analyze nodules at different scales. (related paper)
Estimating the performance of a classifier, without labeled data [WACV 2014 paper]
A very relevant practical issue today is the availability of several algorithms for classification/detection, and several new datasets for testing. There is typically a mismatch between the training and the testing datasets, which make generalization very hard. Even harder is interpreting the benchma
rk performance of the algorithm for the custom testing dataset. We presented an intelligent sampling strategy to work around this issue.