stfxecutables / multichannel-input-pixelwise-regression-u-nets Goto Github PK

This is the official Pytorch implementation of Multichannel input pixelwise regression 3D U-Nets for medical image estimation with 3 applications in brain MRI, submitted at MIDL 2021 as short paper. https://openreview.net/pdf?id=JG895xlWsfA

The U-Net is a robust general-purpose deep learning architecture designed for semantic segmentation of medical images, and has been extended to 3D for volumetric applications such as magnetic resonance imaging (MRI) of the human brain. An adaptation of the U-Net to output pixelwise regression values, instead of class labels, based on multichannel input data has been developed in the remote sensing satellite imaging research domain. The pixelwise regression U-Net has only received limited consideration as a deep learning architecture in medical imaging for the image estimation/synthesis problem, and the limited work so far did not consider the application of 3D multichannel inputs. In this paper, we propose the use of the multichannel input pixelwise regression 3D U-Net (rUNet) for estimation of medical images. Our findings demonstrate that this approach is robust and versatile and can be applied to predicting a pending MRI examination of patients with Alzheimer's disease based on previous rounds of imaging, can perform medical image reconstruction (parametric mapping) in diffusion MRI, and can be applied to the estimation of one type of MRI examination from a collection of other types. Results demonstrate that the rUNet represents a single deep learning architecture capable of solving a variety of image estimation problems.

In application 1 (predicting a pending MRI examination of patients with Alzheimer's disease), We did skull stripping and co-registered each volume to the corresponding target. The following are commands we have used:

skull stripping:

bet <input> <output> -f 0.15 -S -B

Registration (using FreeSurfer):

mri_robust_register --mov <mov.mgz> --dst <dst.mgz> --lta <m2d.lta> --mapmov <aligned.mgz> --iscale --satit --affine 

We use a 5-level 3D U-Net architecture, with Leaky ReLU activation (), learning rate (), Adam optimizer, mean average error (MAE) loss function, z-score intensity normalization and co-registered volumes resized to 128x128x128 for each tasks. Batch size was 3 in applications 1 and 3, and 1 in application 2. We compare all approaches with mean squarederror (MSE), MAE, structural similarity index measure (SSIM), and peak signal to noise ratio (PSNR).

If any of the results in this paper or code are useful for your research, please cite the corresponding paper:

@inproceedings{Wang2021,
   author = {Jueqi Wang and Derek Berger and David Mattie and Jacob Levman},
   journal = {International conference on Medical Imaging with Deep Learning},
   title = {Multichannel input pixelwise regression 3D U-Nets for medical image estimation with 3 applications in brain MRI},
   year = {2021},
}