Abdel-Badeeh M.Salem, Khaled A.Nagaty, Bassant M.El Bagoury

Adaptation in Case-Based Reasoning (CBR) is a very difficult knowledge-intensive task, especially for medical diagnosis. This is due to the complexities of medical domains, which may lead to uncertain diagnosis decisions. In this paper, a new hybrid adaptation model for cancer diagnosis has been developed. It combines transformational and hierarchical adaptation techniques with certainty factors (CF's) and artificial neural networks (ANN's). The model consists of a hierarchy of three phases that simulates the expert doctor reasoning phases for cancer diagnosis, which are Suspicion, To-Be-Sure and Stage phases. Each phase uses the learning capabilities of a single ANN to learn the adaptation knowledge for performing the main adaptation task. Our model first formalizes the adaptation knowledge using IF-THEN transformational rules and then maps the transformational rules into numeric or binary vectors for training the ANN at each phase. The transformational rules of the Suspicion phase encode assigned CF's to reflect the expert doctors' feelings of cancer suspicion. The model is applied to thyroid cancer diagnosis and is tested with 820 patient cases, which are obtained from the expert doctors in the National Cancer Institute of Egypt. Cross-validation test has shown a very high overall diagnosis performance (accuracy rate) that reaches 99.47%. The hybrid adaptation model is described in the context of a prototype namely; Cancer-C that is a hybrid expert system, which integrates neural networks into the CBR cycle.