Dokument: Applications of Supervised Deep (Transfer) Learning for Medical Image Classification
| Titel: | Applications of Supervised Deep (Transfer) Learning for Medical Image Classification | |||||||
| URL für Lesezeichen: | https://docserv.uni-duesseldorf.de/servlets/DocumentServlet?id=59425 | |||||||
| URN (NBN): | urn:nbn:de:hbz:061-20220427-130945-8 | |||||||
| Kollektion: | Dissertationen | |||||||
| Sprache: | Englisch | |||||||
| Dokumententyp: | Wissenschaftliche Abschlussarbeiten » Dissertation | |||||||
| Medientyp: | Text | |||||||
| Autor: | Kronberg, Raphael Marvin [Autor] | |||||||
| Dateien: |
| |||||||
| Beitragende: | Prof. Dr. Markus Kollman [Gutachter] Prof. Dr. Philipp Lang [Gutachter] | |||||||
| Dewey Dezimal-Klassifikation: | 000 Informatik, Informationswissenschaft, allgemeine Werke » 004 Datenverarbeitung; Informatik | |||||||
| Beschreibung: | Abstract:
Medical imaging procedures are central components in the diagnosis of fractures and tumors. Deep Learning, a subfield of Machine Learning, has already established itself in radiology, while other fields such as pathology are also increasingly discovering the power of Deep Learning-based image classification for their own workflows. The automated analysis of many medical images saves time, resources, and money. In medicine, Deep Learning has various applications including the diagnosis of diseases, the more rapid development of drugs, and the personalization of treatments. In the first application of Machine Learning that we include in the field of diagnosis of diseases, we used Deep Transfer Learning to analyze scanned histological hematoxylin and eosin (H&E) stained tissue sections. We showed that our neural network is able to identify and localize pancreatic metastases in healthy lymph nodes. The network can thus be used to assist pathol- ogists with the automated evaluation of numerous tissue sections by having the algorithm pre-filter the data and alert the pathologist to certain sections that require a more detailed investigation. The second application of Machine Learning presented in this thesis is related to the faster development of drugs via Deep Transfer Learning. Using light images of virus-infected cells, we can automatically classify the effectiveness of drugs against a given virus and evaluate the toxicity of the drugs. This approach, which has only been tested under laboratory conditions to date, allows for the rapid, automated analysis of many different drugs. The final application of Machine Learning that we cover is in the area of treatment personalization. When brain tumors are suspected, the protocol includes collecting four MRI sequences (T1, T1CE, T2, and FLAIR). Since patients are often claustrophobic or in poor physical condition, not all four sequences can always be acquired at all. Therefore, we calculated the optimal approach for acquiring the MRI sequences with the maximum information gain as measured by the F1 score of the segmentation neural network and we present a proposal for a shortened acquisition sequence for this type of patient. Our work can be extended in many ways and opens up the possibility of automating time-consuming and cost-intensive processes in clinical routine and basic research in the analysis of medical imaging. | |||||||
| Lizenz: |  Urheberrechtsschutz | |||||||
| Fachbereich / Einrichtung: | Mathematisch- Naturwissenschaftliche Fakultät | |||||||
| Dokument erstellt am: | 27.04.2022 | |||||||
| Dateien geändert am: | 27.04.2022 | |||||||
| Promotionsantrag am: | 01.12.2021 | |||||||
| Datum der Promotion: | 15.03.2022 |
