Abstract

Abstract
Bayesian network structures are usually built using only the data and starting from an empty network or from a naive 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. In this work, we take Bayesian networks built by specialists and show that minor perturbations to this original network can yield better classifiers, while maintaining most of the interpretability of the original network.

Abstract
We evaluated a population of 7199 children between 2 and 19 years old to study the relations between the observed demographic and physiological features in the occurrence of a pathological/non-pathological heart condition. The data was collected at the Real Hospital Portugues, Pernambuco, Brazil, We performed a feature importance study, with the aim of categorizing the most relevant variables, indicative of abnormalities. Results show that second heart sound, weight, heart rate, height and secondary reason for consultation are important features, but not nearly as decisive as the presence of heart murmurs. Quantitatively speaking. systolic murmurs and a hyperphonetic second heart sound increase the odds of having a pathology by a factor of 320 and 6, respectively.

Abstract
Diabetes management is a complex problem. The patient needs to monitor several parameters in order to react in the most appropriate way. Different situations require the diabetic to understand and evaluate different rules. The main source of knowledge for those rules arises from medical practice and is usually transmitted through medical appointments. Given this initial advice, most patient are on a continuous process of managing the disease, toward achieving the best possible quality of life. Motivated by recent aadvances in diabetes monitoring devices, we introduce a diabetes support system designed to accompany the user, advising her and providing early guidance to avoid some of the many complications associated with diabetes. To accomplish this goal, we incorporate standard medical protocols, advice and directives in a Rule Based System (RBS). This RBS which we call Advice Rule Based System (ARBS) is capable of advising and uncovering possible causes for different occurrences. We believe that this solution is not only beneficial to the patient, but may also may be of use to the clinitians advising the patient. The device has continuous contact with the patient, thus it can provide early response if/where needed, Moreover, the system can provide useful data, that an authorized medical expert can use while prescribing a particular treatment, or even when investingating this health problem. We have started to add data-mining algorithms and methods, to uncover hidden behavioural patterns that may lead to crisis situations. Ultimately, through refining the rule systems base don human and machine learning, our approach has the potential for personalising the system according to the habits and phenotype of its user. The system is to be incorporated in a currently developed diabetes management application for Android.

Abstract
Diabetes management is a complex and a sensible problem as each diabetic is a unique case with particular needs. The optimal solution would be a constant monitoring of the diabetic's values and automatically acting accordingly. We propose an approach that guides the user and analyses the data gathered to give individual advice. By using data mining algorithms and methods, we uncover hidden behaviour patterns that may lead to crisis situations. These patterns can then be transformed into logical rules, able to trigger in a particular context, and advise the user. We believe that this solution, is not only beneficial for the diabetic, but also for the doctor accompanying the situation. The advice and rules are useful input that the medical expert can use while prescribing a particular treatment. During the data gathering phase, when the number of records is not enough to attain useful conclusions, a base set of logical rules, defined from medical protocols, directives and/or advice, is responsible for advise and guiding the user. The proposed system will accompany the user at start with generic advice, and with constant learning, advise the user more specifically. We discuss this approach describing the architecture of the system, its base rules and data mining component. The system is to be incorporated in a currently developed diabetes management application for Android.

Abstract
The main goal of this work is to produce machine learning models that predict the outcome of a mammography from a reduced set of annotated mammography findings. In the study we used a dataset consisting of 348 consecutive breast masses that underwent image guided core biopsy performed between October 2005 and December 2007 on 328 female subjects. We applied various algorithms with parameter variation to learn from the data. The tasks were to predict mass density and to predict malignancy. The best classifier that predicts mass density is based on a support vector machine and has accuracy of 81.3%. The expert correctly annotated 70% of the mass densities. The best classifier that predicts malignancy is also based on a support vector machine and has accuracy of 85.6%, with a positive predictive value of 85%. One important contribution of this work is that our model can predict malignancy in the absence of the mass density attribute, since we can fill up this attribute using our mass density predictor.