h1

TY  - THES
AU  - Ghaffari Laleh, Narmin
TI  - Predicting tumor properties from diagnostic pathology whole slide images using weakly supervised deep learning methods
PB  - RWTH Aachen University
VL  - Dissertation
CY  - Aachen
M1  - RWTH-2024-03766
SP  - 1 Online-Ressource : Illustrationen
PY  - 2024
N1  - Veröffentlicht auf dem Publikationsserver der RWTH Aachen University
N1  - Dissertation, RWTH Aachen University, 2024
AB  - Personalized oncology provides cancer patients treatments tailored to their unique genetic and molecular profiles, which can improve clinical outcome and reduce side effects. The adoption of innovative technologies is one of the fundamental tools for refining and advancing personalized oncology. A noteworthy tool within this field is that of deep learning (DL), a facet of artificial intelligence (AI). DL has the potential to revolutionize digital pathology by enhancing diagnosis and extracting clinical biomarkers from standard diagnostic slides. However, before DL can be fully adopted in clinical pathology, we must address challenges such as the need for specific dataset sizes for DL training and concerns over its generalizability and vulnerabilities. This study delves into the capabilities and challenges of DL in the field of computational pathology. In the first part of our study, we utilized DL techniques to detect microsatellite instability (MSI) / mismatch repair deficiency (dMMR) directly from Hematoxylin and Eosin (H</td><td width="150">
AB  - E)-stained whole slide image (WSI) of colorectal cancer (CRC). Our results indicated that the DL-based MSI detectors outperformed clinical-grade performance without requiring any manual annotations. Furthermore, our DL approach could accurately identify CRC patients with microsatellite stability (MSS) in approximately 50
LB  - PUB:(DE-HGF)11
DO  - DOI:10.18154/RWTH-2024-03766
UR  - https://publications.rwth-aachen.de/record/983920
ER  -