基于CT切片的自监督算法识别身体部位 (SSBR).zip

# Self-supervised body part regressor (SSBR) Developed by Ke Yan (ke.yan@nih.gov, [yanke23.com]()), Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, National Institutes of Health Clinical Center If you use this code, please read and cite the following works: * Ke Yan, Le Lu, Ronald Summers, "Unsupervised Body Part Regression via Spatially Self-ordering Convolutional Neural Networks", IEEE ISBI, 2018, https://arxiv.org/abs/1707.03891 * Ke Yan, Xiaosong Wang, Le Lu, Ling Zhang, Adam Harrison, Mohammadhad Bagheri, Ronald M. Summers, "Deep Lesion Graphs in the Wild: Relationship Learning and Organization of Significant Radiology Image Findings in a Diverse Large-scale Lesion Database", IEEE CVPR, 2018, https://arxiv.org/abs/1711.10535 ## Function Predict a **continuous score** for an axial slice in a CT volume which indicates its relative position in the body, e.g. the figure below. The actual correspondence between values and positions needs to be observed when using. See the paper. Samples of unsupervisedly learned body-part scores:  ## Requirements 1. Caffe - use "conda install caffe-gpu" for GPU caffe or "conda install caffe" for CPU 2. easydict - install with "pip install easydict" 3. nibabel (for NIFTI inference) 4. scikit-learn 5. (for training only) VGG-16 pretrained caffemodel (optional, because the algorithm works well even if trained from scratch given enough data). 6. (for training only) Unlabeled training volumes, each volume stored in a folder of 2D slices named by <slice_index>.png. List the names of volume folders in a list file and put the list file's name in TRAIN_IMDB of train.sh. Specify the name of the folder containing all volumes in DATA_DIR of config.yml. If you want to use different data format, change data_layer.py. ## Inference with .nii.gz Run `python nifti_inference.py filename.nii.gz` Results will be written to "slice_scores*.txt". There are two columns - first is raw scores, 2nd is the linear regression fit. If you need to add the -1024 HU offset, set the OFFSET variable in line 99 of nifti_inference.py to -1024 before running. ## Inference with .png To do inference with .png put images in bodypartregressor/test_data/, then run `python bodypartregressor/deploy.py`. ## Installation Minimal packaging of this repository was performed to allow for installation of nifti_inference. To install in your system, run `python setup.py install` from this directory. Alternatively, run `pip install .` Then you can use the NiFTI inference capability of the package as follows: ```python from bodypartregressor.nifti_inference import nifti_inference ctnib = nib.load(file) #read CT using Nibabel ct = ctnib.get_data().astype(np.float32) #get CT as numpy array slice_scores = nifti_inference(ct) #get a list of scores ``` ## Output The slice_scores.txt file obtained using nifti_inference.py now has two columns - the raw score and the linear regressed score. To see a graph showing the linear regression line, uncomment lines 121-125 in nifti_inference.py. ## Changelog Update 7/2020 by Daniel C. Elton: * Converted from Python 2 to Python 3 * added nifti_inference.py, to allow for runs on NiFTI images. * added setup.py to make installable. Update 7/2021 by Daniel C. Elton: * perform linear regression on the output so score vs slice # is a straight line. This smooths out irregularities. Both the raw score and score on the regressed line are outputted. ## Notes * The trained model is in snapshots folder. * When training, see the requirements below and run train.sh. * The provided trained model was trained on 4400 unlabeled CT volumes with various reconstruction filters, scan ranges, and pathological conditions. Random 2D patch cropping was used when training. It is expected to be more accurate in shoulder, chest, abdomen, and pelvis because of the training data. * Input soft tissue window (-175~275 HU) 8-bit images with size 128x128. If your data are different in windowing, image size, scan range etc., it is easy to retrain the algorithm to get a better model for your application. It is also possible to extend the algorithm to sagittal/coronal planes, MR volumes, etc. * The output of SSBR can be used to roughly locate slices of certain body-parts, input to other CAD algorithms as features, detect abnormal volumes, and so on. See paper. ## License MIT License (see LICENCE.md file) We utilize open source code from [py-faster-rcnn](https://github.com/rbgirshick/py-faster-rcnn).