“Improvement of automated coronary calcium scoring and reduction of labeling costs using multi-task deep learning.“
We are happy to announce that our dear colleague Bernhard Föllmer successfully released his first publication. We are hoping that many more will follow and wish you great success!
In our recent work , we could show that joint weak segmentation of coronary artery regions and segmentation of coronary calcifications can improve calcium scoring performance and reduce labeling costs. The multi-task model was trained with an uncertainty-weighted loss function and reached optimal performance with only 12% of the training data. We evaluated our model on patients in three different datasets (DISCHARGE , CADMAN , orCaScore ) and analyzed the influence of various factors such as image noise, metal artifacts, motion artifacts, and image quality. The good performance and reduced number of image slices required pave the way for models that can be used in a clinical setting.
Read more: https://pubmed.ncbi.nlm.nih.gov/35861655/
|Föllmer B, Biavati F, Wald C, Stober S, Ma J, Dewey M, Samek W. Active multi-task learning with uncertainty weighted loss for coronary calcium scoring. Med Phys. 2022 Jul 21. doi: 10.1002/mp.15870. Epub ahead of print. PMID: 35861655.|
|P. Maurovich-Horvat et al., CT or Invasive Coronary Angiography in Stable Chest Pain, New England Journal of Medicine , 1–12 (2022).|
|M. Dewey, M. Rief, P. Martus, B. Kendziora, S. Feger, H. Dreger, S. Priem, F. Knebel, M. Böhm, P. Schlattmann, B. Hamm, E. Schönenberger, M. Laule, and E. Zimmermann, Evaluation of computed tomography in patients with atypical angina or chest pain clinically referred for invasive coronary angiography: randomised controlled trial., BMJ (Clinical research ed.) 355, i5441 (2016).|
|I. Jelmer M. Wolterink, Bob D. de Vos, Tim Leiner, Max A. Viergever, orCaScore, https://orcascore.grand-challenge.org/, 2021.|