University of Twente Student Theses
OOD Detection on Medical Images and Explainable OOD
Imre, Berk (2024) OOD Detection on Medical Images and Explainable OOD.
|
PDF
5MB |
| Abstract: | This research paper investigates various out-of-distribution (OOD) detectors in machine learning (ML) and deep learning (DL), specifically in computer vision in the field of medical images, for which detection of OOD is very crucial. Given that DL models are more popularly used than ML models with state-of-the-art methods, DL-based OOD detection will be the main focus of the research. Systematic experiments will be conducted to analyze the performance and reliability of the OOD detectors on medical images with the increasing range of severity in terms of distribution drift. This paper also introduces an explainability approach to out-of-distribution (OOD) images in DL models to understand the behavior of models, particularly regarding how OOD data affects downstream tasks such as classification and sources of failure of DL models in the presence of OOD data. The findings of this research highlighted the strong performance of PyTorch-OOD in OOD detection in healthcare DL applications. The findings showed the substantial effect of OOD on the confidence and accuracy of DL models in classification tasks. Additionally, noise and misleading visual similarities were identified as the main sources of failure for DL models in classification tasks, in the presence of OOD input. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science BSc (56964) |
| Link to this item: | https://purl.utwente.nl/essays/101095 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page