Samuel Bohumel
Supervisor(s): Ing. Maroš Kollár
Slovak Technical University
Abstract: Neural networks and deep learning are now widely used
approaches for solving tasks in multiple domains, includ-
ing computer vision. In the field of medical image pro-
cessing, these approaches can bring efficient and fast di-
agnosis. However, there is a challenge associated with the
lack of annotated training data needed to train the models.
The collection and especially the annotation of such data
can be time-consuming and expensive. In this work, we
explore the use of generative models for medical data syn-
thesis that could effectively complement existing training
sets and improve the performance of classification models.
The main area of our research is the synthesis of atypical
cells, which is the main signal of a tumor. Nuclear atypia is
usually manifested by enlarged cells and irregular shapes,
which are the features we focus on. We take advantage
of diffusion probabilistic models that are used for guided
synthesis of samples either from a segmentation mask or
an atypia class. This research contributes to the integration
of machine learning techniques in healthcare and evaluates
the presence of synthetic data in training sets.
Keywords: Computer Vision, Image Processing
Full text: Year: 2025
