Nowadays, deep learning techniques are playing an important role in different areas due to the fast increase in both computer processing capacity and availability of large amount of data. Their applications are diverse in the field of bioimage analysis, e.g. for classifying and segmenting microscopy images, for automating the localization of proteins or for automating brain MRI segmentation. Our goal in this project consists in including these deep learning techniques in ImageJ – one of the most used image processing programs in this research community. To do this, we want to develop an ImageJ plugin from which to use the models and functionalities of the main deep learning frameworks (such as Caffe, Keras or Tensorflow). It would be feasible to test the suitability of different models to the problem that is being studied at each moment, avoiding the problems of interoperability among different frameworks. As a first step, we will define an API that allows the invocation of deep models for object classification from several frameworks; and, subsequently, we will develop an ImageJ plugin to make the use of such an API easier.
[1]
Kevin W. Eliceiri,et al.
The ImageJ Ecosystem: An Open and Extensible Platform for Biomedical Image Analysis.
,
2017
.
[2]
Trevor Darrell,et al.
Caffe: Convolutional Architecture for Fast Feature Embedding
,
2014,
ACM Multimedia.
[3]
Thomas Brox,et al.
U-Net: Convolutional Networks for Biomedical Image Segmentation
,
2015,
MICCAI.
[4]
Alfredo Benso,et al.
Beyond Homology Transfer: Deep Learning for Automated Annotation of Proteins
,
2017,
bioRxiv.
[5]
Martín Abadi,et al.
TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
,
2016,
ArXiv.
[6]
Daniel L. Rubin,et al.
Deep Learning for Brain MRI Segmentation: State of the Art and Future Directions
,
2017,
Journal of Digital Imaging.