Fuzzy Data to Crisp Estimates: Helping the Neurosurgeon Making Better Treatment Choices for Stroke Patients

Ischemic stroke of brain manifests itself in the form of loss of blood flow at certain parts of the brain rendering them deprived of oxygen, resulting in a chemical imbalance and death of brain cells in that region. The volume depicted by these cells represents the Infarction volume. This volume defines some of the very sensitive treatment decisions that the neurosurgeon has to make; (a) perform a hemicraniectomy? and (b) Prognosis of this surgery's outcome? Current clinical practice does not provide the surgeons with the answers to the above questions. In this paper, a strategy has been presented that utilizes the Infarction Growth Rate (IGR) as the key element in defining the infarction volume reaching critical levels such that a surgery is inevitable within 48 hours. As a current practice, the stroke lesion growth is most frequently assumed linear, or logarithmic. In this paper, a Machine Learning perspective is presented for mapping the infarction volume using several critical clinical parameters into a possible volumetric prediction in time. The same approach is then used for predicting whether the surgery will be needed soon or not, as well as what might be the likelihood of patient's health in a post-surgery state. In this paper, a machine learning platform is presented which is based on the Adaptive Neuro-Fuzzy Inference System [ANFIS], and has been re-structured such that it can predict IGR and IV with reasonable accuracy, over wide time range. ANFIS hypothesize relationships within the data, and newer learning is able to produce complex characterizations of those relationships. The study was conducted on real stroke-registry database from the local hospital and has shown over 90% accurate prediction.