Breast cancer is the leading cause of cancer-related death for women in Tunisia and the prognosis of its metastasis remains a major problem for oncologists despite advances in treatment. In this work we use Bayesian n...Breast cancer is the leading cause of cancer-related death for women in Tunisia and the prognosis of its metastasis remains a major problem for oncologists despite advances in treatment. In this work we use Bayesian networks to develop a decision support system that is based on the modeling of relationships between key signaling proteins and clinical and pathological characteristics of breast tumors and patients. Motivated by the lack of prior information on the parameters of the problem, we use the Implicit inference for the structure and parameter learning. A dataset of 84 Tunisian breast cancer patients was used and new prognosis factors were identified. The system predicts a metastasis risk for different patients by computing a score that is the joint probability of the Bayesian network using parameters estimated on the learning database. Based on the results of the developed system we identified that overexpression of ErbB2, ErbB3, bcl2 as well as of oestrogen and progesterone receptors associated with a low level of ErbB4 was the predominant profile associated with high risk of metastasis.展开更多
文摘Breast cancer is the leading cause of cancer-related death for women in Tunisia and the prognosis of its metastasis remains a major problem for oncologists despite advances in treatment. In this work we use Bayesian networks to develop a decision support system that is based on the modeling of relationships between key signaling proteins and clinical and pathological characteristics of breast tumors and patients. Motivated by the lack of prior information on the parameters of the problem, we use the Implicit inference for the structure and parameter learning. A dataset of 84 Tunisian breast cancer patients was used and new prognosis factors were identified. The system predicts a metastasis risk for different patients by computing a score that is the joint probability of the Bayesian network using parameters estimated on the learning database. Based on the results of the developed system we identified that overexpression of ErbB2, ErbB3, bcl2 as well as of oestrogen and progesterone receptors associated with a low level of ErbB4 was the predominant profile associated with high risk of metastasis.
