Stroke classification in simulated electromagnetic imaging

• Κύριος συγγραφέας: Πεσλή, Κωνσταντίνα
• Άλλοι συγγραφείς: Pesli, Konstantina
• Γλώσσα: English
• Έκδοση: 2022
• Θέματα: Strokes; Hemorrhage; Ischemia; Electromagnetic simulation; Imaging techniques; Graphs; Machine learning; Classification; Support vector machine; Neural networks; Εγκεφαλικά επεισόδια; Αιμορραγία; Ισχαιμία; Ηλεκτρομαγνητική προσομοίωση; Τεχνικές απεικόνισης; Γράφοι; Μηχανική μάθηση; Ταξινόμηση; Μηχανή διανυσμάτων υποστήριξης; Νευρωνικά δίκτυα

Strokes are ranked internationally as the fifth greatest threat to human health, after heart disease and cancer. Undoubtedly, one of the greatest threats to human health is Covid-19, the scourge of modern times, which seems to be able to cause sudden blood clots, possibly leading to stroke. The increase in pandemic cases combined with the increased incidence of more severe and worse in outcome strokes in otherwise healthy and younger individuals, exacerbates the need for a diligent search for new methods to improve stroke management. The severity of stroke disease is exacerbated if anyone considers the high rates of physical disability and mortality that occur in people regardless the ages. Recent developments in medicine make it possible to treat strokes with better results, but speed remains a key factor in diagnosis and treatment. A typical example of the effort to improve diagnostic tools was the study entitled "Stroke Classification in Simulated Electromagnetic Imaging Using Graph Approaches" which presents a novel graph degree mutual information (GDMI) approach to distinguishing the basic stroke subtypes of: intracranial hemorrhage (ICH) from ischemic stroke (IS), without the need to use expensive brain image reconstruction algorithms that require a lot of time and resources. According to the research that was a source of interest for writing this essay, ICH and IS signals are generated that are simulated using a 16-antenna electromagnetic imaging head system and analyzed for GDMI evaluation. The data collected from each computable human model consists of 256 reflected and received signals. Each signal is converted to a graph to avoid the variable amplitudes of the signal, while for each pair of graphs the relationship between them is calculated by finding the mutual information. These results are forwarded to a support vector machine to determine the stroke subtype. Noise is also injected into the collected signals to test the robustness of the algorithm. The results of the present study show 84% accuracy in detecting ICH from IS and vice versa, while the execution time required to export and classify graph features, depends on the computing capabilities of the computer. Therefore, it can be assumed that by using powerful computer systems to find the key features that distinguish hemorrhagic from ischemic stroke, stroke management can be short enough to meet emergency requirements. In addition, it can be concluded that the detection of subtypes of strokes through the proposed method provides quite optimistic results, as it is considered more economical in contrast to frequency-based electromagnetic imaging systems that use radar or computed tomography algorithms which are computationally expensive.