Publications by authors named "Jean-Francois Samson"
BMC Med Inform Decis Mak
May 2024
Article Synopsis
- - The study addresses challenges in chest X-ray imaging, particularly domain inconsistency during cross-domain transfer learning, by introducing a self-supervised learning strategy called BarlowTwins-CXR to enhance abnormality detection.
- - Using two public datasets (NIH and VinDr-CXR), the study employs a two-stage training approach: self-supervised pre-training followed by supervised fine-tuning, achieving improved diagnostic performance metrics.
- - Results indicate a 3% increase in model accuracy and better localization of abnormalities compared to traditional methods, with training on over 130,000 images, demonstrating the effectiveness of BarlowTwins-CXR in medical imaging.
