Advancing Mammographic Image Generation Using Generative Cellular Automata for Improved CAD Systems

Identifier:  TFM:2522

Handle: https://hdl.handle.net/20.500.11797/TFM2522

Authors:  Rothörl, Lea

Abstract:
This thesis explores applying the Generative Cellular Automata (GeCA) framework for synthesizing mammographic images using the VinDr-Mammo dataset. The framework was adapted for customizable image handling and extended with a breast density classifier to evaluate its impact on a downstream task. Results show GeCA can generate perceptually realistic mammograms and improve macro F1-scores in underrepresented density classes, though binary metrics varied. Limitations include reliance on a single dataset, small image size, and a custom classifier. Still, the study highlights GeCA’s potential for medical image synthesis, warranting further research with diverse datasets and clinical tasks.