Abstract:

Recent works in images mining using deep learning models has yielded state of the art results on a variety of image mining and processing tasks. The deep learning popularity is due to its ability to learn ideal representation of data during the training process without the need of extracting features. It allows systems to learn complex functions that directly map raw sensory input data to the output, without relying on human-crafted features using domain knowledge. Deep learning is watershed for medical images processing. It offers to these field new opportunities in many tasks as: classification, images recognition, segmentation, registration, image retrieval, images generation, etc. The dilemma that confronts many researchers is the starting point: the choice of the adequate deep learning architecture and existing libraries suitable to their finality, as well the availability of several medical datasets. This paper gives interest in multimodal approaches, batch learning, semantic image indexing, and semantic images retrieval which are feebly explored by scientist. We will propose our experience to guide and help researchers all over their evolving learning way. We will give an ample idea of existing deep learning techniques as well medical data sets, software and released works to excerpt considerable open challenges and directions for future researches.