Topic: Out-of-Distribution Detection for Clinical Decision Support in Radiology
Speaker: Alessandro Wollekt, GCB doctoral candidate (Lasser Group)
Date and Time: Tuesday, 15 February 2022, 1:00 pm
Abstract: Deep neural networks show promising results for identifying pathologies on chest radiographs, and therefore could help radiologists improve their clinical routines. However, recent studies have shown that these networks give overconfident predictions for samples out of their training data distribution. In a clinical setting, these models need to handle erroneous input safely. In this presentation, I will highlight the problem of out-of-distribution images and talk about approaches for detecting them.
More information
This semester, the MIBE Seminar will take place online (Zoom). Please register via email (katharina.scholz(at)tum.de) and we are happy to provide the access data.
Further Program Winter Term 2021 / 2022
Our MIBE seminar series in winter term 2021/22 ends with this talk.
It starts again in summer term from May 2022 on.