ExpertBayes

Bayesian network structures are usually built using only the data and starting from an empty network or from a naïve Bayes structure. Very often, in some domains, like medicine, a prior structure is already known based on expert knowledge. This structure can be automatically or manually refined in search for better performance models.

ExpertBayes is a system that implements an algorithm that can refine previously built networks. It allows for (1) reducing the computational costs involved in learning the network structure and parameters only from the data, (2) embedding knowledge of an expert in the newly built network and (3) manual building of fresh new graphical representations. The main ExpertBayes algorithm is random and implements three operators: insertion, removal and reversal of edges. In all cases, source and destination nodes are also chosen randomly. Operators are always applied to the original network, reducing thus the search space and maintaining the graph as close as possible to the intended expert meaning. When used in interactive mode, ExpertBayes allows the user to manually apply any operator and calculate scores on the new model.



MammoClass

The MammoClass tool incorporates machine learning models that predict the outcome of a mammography from a reduced set of annotated mammography findings. The main goal of this application is to help to avoid exposing healthy women to extra surgical or screening procedures. Another goal is the study on the actual relevance of mass density in the findings, since this is one of the attributes that usually is not regarded relevant by physicians. According to physicians, mass density is a feature usually considered to be difficult to annotate, because of the breast tissue and fat composition.

The benefits of an application such as MammoClass is two-fold: a) Radiologists can use MammoClass to confirm the cases they are almost certain of being malignant or benign; b) After getting further confirmation from MammoClass the radiologists can spend more time analyzing the most challenging cases, which are the borderline ones. In this scenario, MammoClass also provides a valuable second opinion that can be incorporated by the radiologists in their process of assessing a mammogram.

MammoClass can be found here.

MammoClass Speech-to-Text (Prototype)

More Information:

MammoClass Poster / Leaflet

MammoClass Supplementary Notes