Artificial Intelligence Methods for Understanding Dynamic Computer Tomography Perfusion Maps

In this article author presents novel approach for analyzing the meaning of brain perfusion maps generated with dynamic computer tomography treatment. With these methods it is possible to detect (if exists), describe position, measure, and state prognosis for brain tissues that are affected by ischemic or hemorrhagic lesions. The whole process is driven by number of image processing algorithms, medical knowledge about average perfusion values and knowledge about interpretation of visualized symptoms. The methods was implemented and tested on 75 triplets of medical images acquired from 30 different adult patients (man and woman) with suspicious of ischemia / stroke. Each triplet was consisted of perfusion CBF and CBV map and ¿plain¿ CT image (one of the image from perfusion treatment acquired before contrast arrival became visible). The algorithm response was compared to image description done to each case by radiologist. The hypothesis to verify was if there is any lesions in perfusion map and if the algorithm found correct position, description and prognosis for them (if the algorithm give a wrong answer for any of this condition the case was considered as ¿error¿). Total error rate (the proportion of error instances to all instances) of full automatic detection (without manual correction of position of brain symmetry axis) was 48.0% and total error rate of semi automatic detection results (with correction of position of brain symmetry axis) was 22.7%.