Histopathology analysis using Vision Transformers

Bc. Tomáš Repák

Diplomová práce

Anotace:

Histopathology is a field of research that analyzes biological tissue. To this day, human experts across the globe manually inspect and analyze samples of biological tissue. This work is of great importance, and as such, it is usually very time-consuming for the human expert. Since a few years ago, methods of Artificial Intelligence, namely convolutional neural networks, have been applied as a tool to help analyze tissue samples and speed up the process. In 2020, a novel neural network model, the Vision Transformer, has been introduced and applied to numerous tasks with great success. In this work, I explore the possibilities of applying Vision Trans- former models and their modifications to histopathological data. Sev- eral modifications of the Vision Transformer architecture are imple- mented and evaluated on two datasets. The first dataset is from the Masaryk Memorial Cancer Institute, and the second is from the Semi- COL Challenge 2023.

Abstract:

Histopathology is a field of research that analyzes biological tissue. To this day, human experts across the globe manually inspect and analyze samples of biological tissue. This work is of great importance, and as such, it is usually very time-consuming for the human expert. Since a few years ago, methods of Artificial Intelligence, namely convolutional neural networks, have been applied as a tool to help analyze tissue samples and speed up the process. In 2020, a novel neural network model, the Vision Transformer, has been introduced and applied to numerous tasks with great success. In this work, I explore the possibilities of applying Vision Trans- former models and their modifications to histopathological data. Sev- eral modifications of the Vision Transformer architecture are imple- mented and evaluated on two datasets. The first dataset is from the Masaryk Memorial Cancer Institute, and the second is from the Semi- COL Challenge 2023.

Keywords

histopathology, tissue, neural networks, transformers, vision transformers, masked autoencoder

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 16. 5. 2023

Obhajoba závěrečné práce

• Obhajoba proběhla 21. 6. 2023
• Vedoucí: doc. RNDr. Tomáš Brázdil, Ph.D.
• Oponent: doc. RNDr. David Svoboda, Ph.D.

Masarykova univerzita

Fakulta informatiky

Magisterský studijní program / obor:
Umělá inteligence a zpracování dat / Strojové učení a umělá inteligence