Stochastic and heuristic modelling for analysis of the growth of pre-invasive lesions and for a multidisciplinary approach to early cancer diagnosis

To this day, lung cancer remains the leading cause of all cancer deaths for both sexes. Current treatment options lead to a cure in only about ten percent of diagnosed cases of lung cancer. One of the main reasons why this type of cancer has such poor prognosis is that it is very difficult to diagnose at the early stages. It is well known that the survival rates can be improved by the early detection of pre-invasive lesions, which are believed to be the possible precursors to malignant tumours. Although new diagnostic devices are allowing numerous lesions to be detected early, it is becoming clear that only a small percentage of these will actually progress to cancer. Therefore, the critical question is how to determine the factors that will define which of these lesions will become malignant. In this thesis, two computational models and a novel approach to represent biological knowledge for use in the early diagnosis of cancer are presented. In the first part, a stochastic model representing the early development of pre-invasive neoplastic bronchial epithelial lesions as contact processes is introduced. The results of the simulations run on this model gave us some insight on the probability of growth of specific lesions. Yet, it also shed light on the fact that for an effective diagnostic tool we would need to consider a lot more information about the patients and their condition beyond the structural behaviour of independent lesions. This led to the development of a new approach to multidisciplinary biological knowledge representation: the Probabilistic Property-Based Model (PPBM). Based on a cognitive model of knowledge construction, PPBM presents a heuristic approach to diagnosis by taking into account multiple-domain elements such as imaging, serum, sputum, cytological and genetic data as well as personal medical history and lifestyle factors.