Neural Network Approach for Identification of Selected Brain Perfusion Abnormalities

In this article the authors propose the neural network (NN) classifier for identification of brain perfusion abnormality type and its localization in CBF and CBV dynamic brain perfusion maps. The approach is based on comparison of average values of CBF or CBV (obtained for healthy brain) with mean CBF or CBV values measured in symmetric regions of interests (ROI) in left and right hemisphere for diagnosed patient. The proposed NN was validated on set of 31 CBF and CBV medical images acquired from 30 different adult patients (man and woman) with suspicious of is chemia / stroke. The use of NN enables not only to recognize the localization and type of perfusion abnormality but also to quantify the uncertainty of automatic diagnosis.