Loading...

Domain Generalization in Deep Learning Models for Histopathology

Sadeghi, Reyhaneh | 2023

19 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56776 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rohban, Mohammad Hossein
  7. Abstract:
  8. Domain shift is an inevitable issue when histopathological images are analyzed in a standard laboratory. This is due to the variations in tissue handling, manual procedures for sample preparation, and differences in scanners. This can result in reduced performance of machine learning algorithms trained on images from one laboratory when applied to another. In this framework, the goal of utilizing domain generalization techniques in machine learning is to develop models that perform well in different domains. This research examines various methods of dealing with the domain shift challenges within the context of detecting mitotic cells, and proposes algorithms to improve domain generalization and detection accuracy. The MIDOG2021 challenge data is used to evaluate the proposed models. Also, the detection rate on out-of-domain images is used as the evaluation criterion. The histopathological data is first generated by using the available MIDOG2021 challenge data. Then, the effectiveness of the RetinaNet network on the generated data is assessed. Our experimental results indicate that the performance of the network is reduced when the inner-domain differences are significant and a small data set is available for training of the network. To address this issue, data augmentation and stain normalization methods are adopted and their impacts on the domain generalization problem are investigated. The experimental results demonstrate that using a Fourier domain adaptation data augmentation method and a generative adversarial network for color normalization, the domain generalization capability is enhanced and the efficiency of the RetinaNet network is improved for the MIDOG2021 challenge data, ultimately achieving an F1 score of 63.35 on the test dataset
  9. Keywords:
  10. Deep Learning ; Histopathology Images ; Domain Generalization ; Domain Shift ; Machine Learning ; Histological Images Analysis