Projects Last Projects

Survival analysis in breast cancer is crucial for personalizing treatment, and deep
learning models applied to histology images offer a promising avenue for improving
prognostic predictions. Tissue Microarrays (TMAs) provide a high-throughput
method for analyzing large patient cohorts, making them a natural data source for
training such models. In this work, we adopt a pipeline architecture using the pre-
trained Titan foundation model to encode entire TMA slides, which are then fed into
a regression head to predict a patient-specific risk score. Our findings demonstrate
that this approach can effectively stratify patients into distinct prognostic groups,
achieving a Concordance Index of 0.6175 on the BCOU dataset. The outcomes are
competitive with other TMA-based models.

Please, see project report.

Please, see final presentation.