Note: There is a more versatile and user-friendly implementation available now
This repository contains code samples and documentation for regression with convolutional neural networks on medical images, described in the following publications:
"Identifying morphological indicators of aging with neural networks on large-scale whole-body MRI" [1]
"Large-scale biometry with interpretable neural network regression on UK Biobank body MRI" [2]
"Large-scale inference of liver fat with neural networks on UK Biobank body MRI" [3]
Contents:
- PyTorch code for network models, training and inference
- Formatting of UK Biobank neck-to-knee body MRI (into volumes and 2D formats)
- Parameters for registration of UK Biobank neck-to-knee MRI with [4]
- Uncertainty code (work in progress)
The code contains the old network configuration [1] in comments, but by default uses the new, optimized hyperparameters and learning policy [2]. It also contains code for the dedicated liver fat fraction measurement [3], including a trained snapshot, which can be found here.
The saliency aggregation is currently not included. We used a modified GitHub repository by Utku Ozbulak, which implements guided gradient-weighted class activation maps [5].
Please note that the UK Biobank data used in the publication can not be made publically available. However, the calculated reference values and split IDs used for the experiments have been shared as return data of application 14237 with the UK Biobank, so that reproducing the results should be possible.
For any questions and suggestions, feel free to reach out!
If you use this code for any derived work, please consider citing [2] and linking this GitHub.
[1] T. Langner, J. Wikström, T. Bjerner, H. Ahlstrom, and J. Kullberg, “Identifying morphological indicators of aging with neural networks on large-scale whole-body MRI,” IEEE Transactions on Medical Imaging, pp. 1–1, 2019.
[2] T. Langner, H. Ahlström, and J. Kullberg, “Large-scale biometry with interpretable neural network regression on UK Biobank body MRI,” Scientific reports, 10.1 (2020): 1-9.
[3] T. Langner, R. Strand, H. Ahlström, and J. Kullberg, “Large-scale inference of liver fat with neural networks on UK Biobank body MRI,” International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCA). Springer, Cham, 2020.
[4] S. Ekström, F. Malmberg, H. Ahlström, J. Kullberg, and R. Strand, “Fast Graph-Cut Based Optimization for Practical Dense Deformable Registration of Volume Images,” arXiv:1810.08427 [cs], Oct. 2018. arXiv: 1810.08427
[5] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization,” in 2017 IEEE International Conference on Computer Vision (ICCV), (Venice), pp. 618–626, IEEE, Oct. 2017.