Bibliogr. s. 48-49

Background: Computer aided diagnosis (CAD) becomes one of the most important diagnostic tools for urgent states in cerebral stroke and other life-threatening conditions where time plays a crucial role. Routine CT is still diagnostically insufficient in hyperacute stage of stroke that is in the therapeutic window for thrombolytic therapy. Authors present computer assistant of early ischemic stroke diagnosis that supports the radiologic interpretations. A new semantic-visualization system of ischemic symptoms applied to noncontrast, routine CT examination was based on multiscale image processing and diagnostic content estimation. Material/Methods: Evaluation of 95 sets of examinations in patients admitted to a hospital with symptoms suggesting stroke was undertaken by four radiologists from two medical centers unaware of the final clinical findings. All of the consecutive cases were considered as having no CT direct signs of hyperacute ischemia. At the first test stage only the CTs performed at the admission were evaluated independently by radiologists. Next, the same early scans were evaluated again with additional use of multiscale computer-assistant of stroke (MulCAS). Computerized suggestion with increased sensitivity to the subtle image manifestations of cerebral ischemia was constructed as additional view representing estimated diagnostic content with enhanced stroke symptoms synchronized to routine CT data preview. Follow-up CT examinations and clinical features confirmed or excluded the diagnosis of stroke constituting 'gold standard' to verify stroke detection performance. Results: Higher AUC (area under curve) values were found for MulCAS -aided radiological diagnosis for all readers and the differences were statistically significant for random readers-random cases parametric and non-parametric DBM MRMC analysis. Sensitivity and specificity of acute stroke detection for the readers was increased by 30% and 4%, respectively. Conclusions: Routine CT completed with proposed method of computer assisted diagnosis provided noticeable better diagnosis efficiency of acute stroke according to the rates and opinions of all test readers. Further research includes fully automatic detection of hypodense regions to complete assisted indications and formulate the suggestions of stroke cases more objectively. Planned prospective studies will let evaluate more accurately the impact of this CAD tool on diagnosis and further treatment in patients suffered from stroke. It is necessary to determine whether this method is possible to be applied widely.