Breast cancer risk prediction with transformers
Title: Breast cancer risk prediction with transformers
SNIC Project: Berzelius-2022-140
Project Type: LiU Berzelius
Principal Investigator: Kevin Smith <ksmith@kth.se>
Affiliation: Kungliga Tekniska högskolan
Duration: 2022-06-30 – 2023-01-01
Classification: 10207
Keywords:

Abstract

The aim of this project is to develop and train deep learning models to recognise anatomic signs of a predisposition for breast cancer by analysing mammographic screening images. Breast cancer research within the context of deep learning has traditionally pertained primarily to CNNs. Significant findings have been achieved regarding particularly breast cancer detection, risk estimation and masking prediction. Recently, however, vision transformers have attracted attention due to their inherent benefits compared with CNNs. Consequently, investigating their performance for the development of breast cancer models is of contemporary relevance. Early detection of breast cancer is often a prerequisite for successful treatment and proficient deep models can assist in accomplishing that. Cancer detection, risk estimation and masking prediction are of circumstantial importance and will subsequently be the subject of this study through the application of vision transformers. In this project, we aim to improve our current breast cancer risk predictors and get new state-of-the-art at the end of the project period.