Neural Networks and Artificial Intelligence

SYNOPSIS Assessment of surgical risk in patients undergoing pulmonary resection is a fundamental goal for thoracic surgeons. Usually used risk scores do not predict the individual outcome. Data Mining (DM) and Artificial Neural Networks (ANN) are artificial intelligence mathematical models that have been used for estimation of several prognostic situations. When used to assess surgical risk they can integrate results from multiple data, predicting the individual outcome for patients rather than assigning them to less precise risk group categories. INTRODUCTION Even though therapeutic strategies look as the main goal of modern medical activity, diagnosis and prognosis keep their traditional central place in patient's care process. Particularly, prognosis implies not only the usefulness of therapy at large but also the impact of the therapeutic act in the immediate outcome of the patient. When surgical procedures are planned, the immediate prognosis in terms of morbidity and mortality is known as surgical risk. An increasingly older patient population with more comorbidities makes the evaluation of surgical risk a relevant everyday activity. The description of different indexes or scores has been an attempt to objectively quantify surgical risk 1 2 , avoiding the uncertain " personal experience " that can only play a partial role in patient's prognosis. Indeed risk indexes can only give a statistic probability of outcome, i.e.: the patient belongs to a certain group with a known percentage of morbidity and/or mortality 3. But this is not exactly an individual prognosis, as it is not the answer to the question of patients and families looking for a personal estimation. The main goal of any system measuring surgical risk should be to accurately know in which point of the prognostic curve each patient is situated. The risk of developing complications or dying could be evaluated by multivariate statistical analysis (such as multiple logistic regression) designed to compare population groups and the relative contribution of varied risk factors to outcome. Logistic regression analysis is applicable to problems that have binary solution (yes/no) and provides likelihood ratios for each population group 4 , but it cannot give a qualitative evaluation for a given individual patient. Artificial Intelligence (AI) can be the clue to obtain this kind of individual prognosis. In fact, different tools (Data Mining, Artificial Neural Networks) are able to " learn " from previous series of cases and can then give the specific answer for the new one. The ability to manage …