Skip to content

Latest commit

 

History

History
29 lines (17 loc) · 930 Bytes

README.md

File metadata and controls

29 lines (17 loc) · 930 Bytes

HeartSegmentation

Testing of several known deep learning models for IRM heart segmentation and a novel model, Double attention Unet.

Double Attention Unet architecture

Inspired by :

  • Oktay O, Schlemper J, Folgoc L L, et al. Attention u-net: Learning where to look for the pancreas[J]. arXiv preprint arXiv:1804.03999, 2018. 

  • D. Jha, M. A. Riegler, D. Johansen, P. Halvorsen et H. D. Johansen, « DoubleU-Net: A Deep Convolutional Neural Network for Medical Image Segmentation », 2020 IEEE 33rd International Symposium on Computer-Based Medical Systems (CBMS),2020, pp. 558-564, doi: 10.1109/CBMS49503.2020.00111.

Results

Learning of Double Attention Unet

Confusion matrix

Tested model stats

Visualisation

Sometimes better than the actual label