<html><head></head><body>Für alle virtuell interessierten, hier der Link zum BBB: <a href="https://conf.fmi.uni-leipzig.de/b/van-bwb-jil-vk2">https://conf.fmi.uni-leipzig.de/b/van-bwb-jil-vk2</a><br><br>(Verbindung im Raum ist schlecht d.h. es kann zum Abbruch oder Unterbrechungen kommen)<br><br>Viele Grüße<br>Vanessa Kretzschmar <br><br><div class="gmail_quote">On October 5, 2022 11:06:24 AM GMT+02:00, Vanessa Kretzschmar <kretzschmar@informatik.uni-leipzig.de> wrote:<blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">
<pre dir="auto" class="k9mail">E I N L A D U N G<hr>zum Forschungsseminar 'Computergrafik, Bildverarbeitung und Visualisierung'<br><br> am Mittwoch, den 12. Oktober 2022, 13:15 Uhr,<br> im Raum SG 3-10 im Seminargebäude an der Universitätsstraße.<hr>Wir hören einen Vortrag von<br><br> Lucas Peter<br><br>mit dem Titel:<br><br> 'Predicting Stroke Lesions from CT Imaging and Clinical Information using UNets'<br><br>zum Inhalt:<br><br> Strokes are one of the leading causes of disability worldwide. The<br> resulting disabilities are caused by regions of abnormal brain tissue<br> that are a result of a blockage in the blood circulation of the brain.<br> To accurately and effectively treat these abnormalities, a precise<br> prediction of where the lesion will form is of huge benefit, as medical<br> personnel would be able to more effectively determine which parts of the<br> brain need to be recanalized. One approach to generate these predictions<br> that has gained popularity recently, has been to use Artificial Neural<br> Networks, that are trained on Magnetic resonance imaging (MRI)- or<br> computer tomography (CT)-Scans of patients. Such an approach is presented<br> here in this work. Specifically, a type of Convolutional Neural Network<br> called a UNet is trained on a comparatively large dataset and different<br> values for multiple hyperparameters are tested and their efficacy is<br> compared. In doing so, a model is trained that generates results that<br> are comparable to the state-of-the-art in multiple performance metrics.<br> Furthermore, the importance of the used imaging and auxiliary clinical<br> modalities is determined and a novel technique to generate the salvageable<br> tissue is discussed.<hr>Alle Interessierten sind im Namen von Professor Dr. Scheuermann herzlich<br>eingeladen.<br><br>Mit freundlichen Grüßen<br>Vanessa Kretzschmar</pre></blockquote></div><div style='white-space: pre-wrap'><div class='k9mail-signature'>-- <br>Sent from my Android device with K-9 Mail. Please excuse my brevity.</div></div></body></html>