Publications by authors named "Mateo Mikulic"
J Imaging Inform Med
July 2024
Article Synopsis
- Multiple studies indicate convolutional neural networks (CNNs) can effectively predict medical conditions, sometimes outperforming human professionals, but they often operate as "black boxes."
- This research investigates how occluding confounding variables in medical images affects model predictions, focusing on osteopenia using the GRAZPEDWRI-DX dataset.
- While models trained on non-occluded images generally performed better in numerical evaluations, radiologists preferred models with occluded images, highlighting a trade-off between model performance and interpretability.
