Data repository for the 2022 AAPM deep-learning spectral CT grand challenge

This repository contains all the materials necessary for training deep-learning based algorithmsfor the DL-spectral CT Challenge.The training data folder contains 1000 cases with which to train your networks.For more information about the spectral CT modeling, please see starting_kit.tgz.

Data are in python numpy's .npy format, and arrays are stored in float32 (single precision).The .npy files can be read into python with the numpy.load command, yielding single precisionfloating point arrays of the proper dimensions.

Files:Phantom_[Tissue].npy.gz   ([Tissue] can be Adipose, Fibroglandular, or Calcification)These arrays are 1000x512x512. Files are gzipped for faster data transfer.1000 images of pixel dimensions 512x512.These are ground truth Tissue maps for the simulated breast phantom.

[low/high]kVpTransmission.npy.gzThese arrays are 1000x256x1024The represent 1000 cases of the low and high kVp transmission data, each of which has 256 projectionsonto a linear 1024-pixel detector.

[low/high]kVpImages.npy.gzThese arrays are 1000x512x5121000 images of pixel dimensions 512x512, generated from [low/high]kVpTransmission.npyby standard negative logarithm processing followed by filtered back-projection (FBP)image reconstruction.

For the challenge, the goal will be to estimate (or predict) Phantom_[Tissue].npyfrom either [low/high]kVpTransmission.npy or [low/high]kVpImages.npy or both.

Iterative or model-based approaches will necessarily use [low/high]kVpTransmission.npy as input.Deep-learning approaches can be:solely image-to-image; i.e. [low/high]kVpImages -> Phantom[Tissue] prediction.solely data-to-image; i.e. [low/high]kVpTransmission -> Phantom[Tissue] prediction.Or some combination of the two.Hybrid iterative/deep-learning approaches are also acceptable.

Participants using image-to-image approaches do not need to know the spectral CT model.All other approaches will need this knowledge and model specifications areprovided in the folder, which has been packed in the gzipped tar-file starting_kit.tgz.

Challenge report is published in Medical Phyiscs:

Sidky EY, Pan X. Report on the AAPM deep-learning spectral CT Grand Challenge. Med Phys. 2024; 51: 772–785.

See link below in Related Works.