A multi-resolution CNN to predict cancer histopathologic features
Panoptes for cancer H&E image prediction and visualization (PyPI version)
panoptes is a InceptionResnet-based multi-resolution CNN architecture for cancer H&E histopathological image features prediction. It is initially created to predict visualize features of endometrial carcinoma (UCEC), hoping to automate and assist gynecologic pathologists making quicker and accurate decisions and diagnosis without sequencing analyses. Details can be found in the paper: https://www.biorxiv.org/content/10.1101/2020.02.25.965038v3 It can also be applied to other cancer types.
Currently, it includes training/validating/testing of following features of endometrial cancer:
- 18 mutations (ARID1A, ATM, BRCA2, CTCF, CTNNB1, FAT1, FBXW7, FGFR2, JAK1, KRAS, MTOR, PIK3CA, PIK3R1, PPP2R1A, PTEN, RPL22, TP53, ZFHX3)
- 4 molecular subtypes (CNV.H, CNV.L, MSI, POLE). if you want to predict only 1 subtype (eg. CNV.H), then type "CNV.H".
- Histological subtypes (Endometrioid, Serous)
- train: training a new model from scratch.
- validate: load a trained model and validate it on a set of prepared samples.
- test: load a trained model and apply to a intact H&E slide.
PC are architectures with the branch integrating BMI and age; P are original Panoptes
- Panoptes1 (InceptionResnetV1-based; P1/PC1)
- Panoptes2 (InceptionResnetV2-based; P2/PC2)
- Panoptes3 (InceptionResnetV1-based; P3/PC3)
- Panoptes2 (InceptionResnetV2-based; P4/PC4)
- Install the package version through pip
pip install panoptes-he
- Requirements are listed in
from panoptes.execute import panoptes, get_default_data
- Get default label_file and split_file:
label_file, split_file = get_default_data()
panoptes(mode, outdir, feature, architecture, log_dir, image_dir, tile_dir=None, modeltoload=None, imagefile=None, batchsize=24, epoch=100000, resolution=None, BMI=np.nan, age=np.nan, label_file=None, split_file=None, cancer='UCEC')
- mode(required): select a mode to use (train, validate, test)
- outdir(required): name of the output directory
- feature(required): select a feature to predict (histology, subtype, POLE, MSI, CNV.L, CNV.H, ARID1A, ATM, BRCA2, CTCF, CTNNB1, FAT1, FBXW7, FGFR2, JAK1, KRAS, MTOR, PIK3CA, PIK3R1, PPP2R1A, PTEN, RPL22, TP53, ZFHX3)
- architecture(required): select a architecture to use (P1, P2, P3, P4, PC1, PC2, PC3, PC4)
- log_dir(required): directory for log file, which will also contain the outdir.
- image_dir(required): directory contains the svs/scn scanned H&E slides
- tile_dir(required for train and validate): directory contains tiles of svs/scn scanned H&E slides
- modeltoload(required for validate and test): full path to trained model to load (without .meta)
- imagefile(required for test): the svs/scn scanned H&E slide to be tested in image_dir
- batchsize(optional, default=24): batch size
- epoch(optional, default=100): max epoch; early stop is enabled.
- resolution(optional): resolution of scanned H&E slides. If known, enter 20 or 40. Otherwise, leave blank.
- BMI(optional): patient BMI for test with PC models.
- age(optional): patient age for test with PC models.
- label_file(required for train and validate): label dictionary. Example can be used
- split_file(optional): For train and validate mode, random split is default for data separation.
- cancer(currently optional): Currently only UCEC is supported.
If you prefer a customized split, please provide a split file (example can be used
Main repository and commandline, bash, GUI version can be found in: https://github.com/rhong3/Panoptes
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size panoptes_he-0.1.16-py3-none-any.whl (275.5 kB)||File type Wheel||Python version py3||Upload date||Hashes View|
|Filename, size panoptes-he-0.1.16.tar.gz (270.4 kB)||File type Source||Python version None||Upload date||Hashes View|
Hashes for panoptes_he-0.1.16-py3-none-any.whl