diff --git a/README.md b/README.md index 96f8773..9c4db02 100644 --- a/README.md +++ b/README.md @@ -1,22 +1,38 @@ # About -Codes for paper **Automatic Knee Osteoarthritis Diagnosis from Plain Radiographs: A Deep Learning-Based Approach.** +The software in this branch implements an automatic pipeline for osteoarthritis severity assessment from plain radiographs. +It can be utilized via REST over HTTP or DICOM protocols. Furthermore, we have built a web UI that is easy to use. -*Tiulpin, A., Thevenot, J., Rahtu, E., Lehenkari, P., & Saarakkala, S. (2018). Automatic Knee Osteoarthritis Diagnosis from Plain Radiographs: A Deep Learning-Based Approach. Scientific reports, 8(1), 1727.* +The core Machine Learning pipeline is backed up by two papers: +1. Tiulpin, Aleksei, Iaroslav Melekhov, and Simo Saarakkala. +"KNEEL: Knee Anatomical Landmark Localization Using Hourglass Networks." +Proceedings of the IEEE International Conference on Computer Vision Workshops. 2019. +2. Tiulpin, Aleksei, et al. +"Automatic knee osteoarthritis diagnosis from plain radiographs: A deep learning-based approach." +Scientific reports 8.1 (2018): 1727. -# Disclaimer +# Authors +The whole package was built by Aleksei Tiulpin (@lext). Egor Panfilov (@soupault) +has built most of DeepKnee v1: https://github.com/MIPT-Oulu/DeepKnee-web (RIP). +The current version of DeepKnee uses only the UI from the previous version. -**This branch is only for inference purposes. Re-training is possible only in the master branch!!!** +# Disclaimer +This software is not a medical device and intended for the use in research settings only. ### Running the software -This code requires the fresh-most docker and docker compose installed. +This code requires the fresh-most docker and docker compose installed. +Be careful, this app carries all the dependencies and weighs around 20GB in total. -Execute `sh deploy.sh cpu` to deploy the app on CPU. If you have installed nvidia-docker, -you can also deploy on GPU. The inference is 3 times faster on GPU. To deploy on GPU, run `sh deploy.sh gpu`. +Execute `sh deploy.sh cpu` to deploy the app on CPU. If you have installed `nvidia-docker`, +you can also deploy on GPU. The inference is 3 times faster on GPU. To deploy on GPU, run `sh deploy.sh gpu`. +To test the software, just check in your browser `localhost:5003`. Here is an example of the result of processing +a single knee X-ray via UI: -Be careful, this app carries all the dependencies and weighs around 10GB in total. +UI pic -# Technical documentation +A YouTube Video describing the process of using the UI is shown here: https://youtu.be/4wcnmsxvim8. +The video has been recorded for DeepKnee v1, but nothing has really changed since then. +# Technical documentation The software is currently composed of six separate loosely-coupled services. Specifically, those are: 1. `KNEEL` - Knee joint and landmark localization (https://arxiv.org/abs/1907.12237). REST microservice, port 5000. @@ -26,7 +42,7 @@ The software is currently composed of six separate loosely-coupled services. Spe 5. `Change polling` - A service that tracks what came to Orthanc and then forwards those data to DeepKnee as well as to PACS where user wants to store the results. By default, we use an embedded orthanc PACS as a remote PACS. However, this store is not persistent and will be emptied upon restart. It is highly recommended to specify a persistent remote PACS -6. `UI` - User Interface implemented in ReactJS. This part runs on 5003. +6. `UI` - User Interface implemented in ReactJS. This part runs on 5003, however the port can be changed in docker compose files. The platform is designed so that it is possible to use `KNEEL` and `DeepKnee` separately. Both microservices expect @@ -35,6 +51,10 @@ it needs to be done to `/kneel/predict/bilateral` or `/deepknee/predict/bilatera An example script that uses the platform can be found in the file `analyze_folder.py`. +Below you can see a schematic representation of the software architecture +DeepKnee Architecture + + ## A few words about PACS integration To deploy this software in your network with persistent PACS, you need to modify docker-compose file which is used to run DeepKnee. Specifically, you need to change the entry point parameters of `dicom-router` service @@ -58,4 +78,4 @@ The provided software is not cleared for diagnostic purposes. year={2018}, publisher={Nature Publishing Group} } -``` +``` \ No newline at end of file diff --git a/pics/deepknee-architecture.png b/pics/deepknee-architecture.png new file mode 100644 index 0000000..be1b17f Binary files /dev/null and b/pics/deepknee-architecture.png differ diff --git a/pics/deepkneeui.png b/pics/deepkneeui.png new file mode 100644 index 0000000..9b506a5 Binary files /dev/null and b/pics/deepkneeui.png differ