Prediction of Sickle Cell Anemia Patient's Response to Hydroxyurea Treatment Using ARTMAP Network

This paper discusses the application of the newly developed Mahalanobis distance-based ARTMAP (MART) network to the predication of sickle cell anemia patients’ response to Hydroxyurea treatment. Mahalanobis distance-based ARTMAP is a modified version of the Fuzzy ARTMAP networks where the activation and choice function are the same and equal to the Mahalanobis distance between the pattern and the template. A recurrent formula has been derived to update the inverse of a covariance matrix after it resonates with an input pattern. The full covariance matrices provide a better coverage of samples than the diagonal covariance matrices. The first part of the paper describes the MART network, with a short introduction of fuzzy ARTMAP and Gaussian ARTMAP networks, from which it was derived. The second part of the paper describes the clinical data and the procedures for data preprocessing, training and testing. The classification accuracy achieved with MART classifier was as high as 92.59%