Institutional Repository of Vilnius University
Multi-task learning for survival analysis using pathology images /
Abstract
dc:descriptionThis thesis introduced a multitask learning framework for survival analysis from a combination of histopathological images and clinical data. The integration of structured tabular data and unstruc tured wholeslide images (WSIs) improves the prediction of survival probability and timetoevent outcomes. The approach is based on usafe of pretrained convolutional neural networks for feature extraction, watershed segmentation for preprocessing, parametric tile selection, and a multilayer perceptron model for multitask inference. The pipeline achieves a significant improvement in sur vivability estimation, reaching concordance index of 0.829 on the validation dataset and 0.823 on the test dataset, and outperforming baseline methods like Cox regression which was analyzed for the same dataset in the previous work. This research contributes to the field by presenting an effi cient methodology for integrating clinical and histopathological data, proposing framework for image preprocessing and tile selection, and employing customized loss functions for survival analysis tasks.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kuzminas, Ovidijus,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD229583048&prefLang=en_US