|
Current machine learning techniques are able to achieve spectacular results in automatic understanding of natural images whereas in the area of medical image analysis the progress is not that evident. The problem is medical knowledge essential for proper interpretation of image content. That knowledge, possessed by relatively small number of radiological or biomedical experts, usually cannot be directly expressed using mathematical formulas. This can be overcome by laborious knowledge acquisition or by techniques to some extent imitating expert behaviour. Both approaches are, however, still challenging tasks. That is why the goal of the special session is to discuss the problems in acquisition and utilization of domain knowledge in automatic understanding of semantic image structure.
Both computer scientists as well as radiological and biomedical experts are welcome as participants. The session should constitute a perfect forum to express expectations, suggest solutions and share experience for members of those communities.
The scope of the session contains, but is not limited to, the following topics:
- deep architectures and learning in image analysis (e.g. convolutional neural networks, LSTM networks, etc.);
- expert knowledge acquisition and representation methods (how effectively medical knowledge can be acquired and used in existing models of image analysis);
- classical image segmentation and object localization techniques capable of using domain specific knowledge (e.g. active contours, neural networks, etc.);
- structural image representation and analysis (e.g. image decomposition, structured prediction, probabilistic graphical models).
|
