Deep learning architecture for data mining from surgical data

The paper addresses issues about data mining from surgical data. Surgical data has several distinguishing features - it is voluminous, heterogeneous, noise-prone, has low level of formalization due to lack of standardization in domain and highly correlated features. Architecture for data mining from surgical data, which deals with described features, is proposed. It is related to deep learning architectures as it consists of several hierarchical levels. Context-wise input layer modules process heterogeneous data. Results from these modules are calibrated before being passed to the next layer. At the last layer inference model for classification and prediction is derived.