The development of competitive intelligent computational means to assist medical decision-making is a matter of high necessity. In this respect, evolutionary algorithms are very appropriate for the given task, due to their high flexibility and generality. Present paper proposes an evolutionary technique with multiple populations that cooperate in order to predict the diagnosis of patients suspected of breast cancer. The method trains on a set of patients that are described by their medical data referring to the cytological information of this type of cancer. Experimental results demonstrate that proposed technique provides an accurate means of understanding the factors that lie behind the diagnosis of breast cancer and a viable way of checking the consistency of judgment.