Niffler: A DICOM Framework for Machine Learning and Processing Pipelines

Niffler is an open source DICOM framework for machine learning and processing pipelines. This is an introductory presentation for Niffler.

 

Niffler: A DICOM Framework for Machine Learning and Processing Pipelines. from Pradeeban Kathiravelu, Ph.D.

[SIIM CMIMI20] A DICOM Framework for Machine Learning Pipelines against Real-Time Radiology Images

Today I presented our Niffler open-source framework at SIIM CMIMI 2020 at ML Algorithms & Toolkits + Infrastructure to Support ML session. We had slightly more than 100 participants, including the 90+ attendees and 10 panelists. We had 8 minutes presentations followed by a unified Q&A session. My presentation slides are given below.

A DICOM Framework for Machine Learning Pipelines against Real-Time Radiology Images from Pradeeban Kathiravelu, Ph.D.

 
This is my virtual conference experience where I presented a paper. I have attended several virtual conferences such as KDD, EuroSys, and DisCoTec this year, thanks to COVID19 lock-down. I also presented my work at IEEE SDS this year. However, SDS was a recording. We already had pre-recorded the videos and shared. As such, CMIMI becomes my first virtual conference experience where I was also a presenter.
 

I miss attending conferences in-person. The virtual conferences do not feel the same at all. Yes, we still see the presentations. But I value that coffee-hour talks, going out with newly made friends/colleagues, finding potential collaborators, personal connections, and many more. But something is better than nothing. We could still watch and listen to the presentations and interact with the presenters. But the face-to-face communication -- I miss that. 

 

I hope we will overcome the pandemic together and be back to in-person or even better - hybrid conferences soon! KDD 2021 has planned to be hybrid, online as well as on-site in Singapore. I am optimistic! The 2 days with CMIMI were great. I wish success to everyone in their research.

Niffler: A DICOM Framework for Machine Learning Pipelines against Real-Time Radiology Images

I have been developing Niffler for quite some time. It is a DICOM network framework for machine learning pipelines. It retrieves the DICOM images from the PACS real-time and retrospectively based on the queries. It has been running for 19 months in our lab stable now, and has powered several machine learning research.

It extracts and stores metadata from the PACS in real-time, and also retrieves studies retrospectively for each specific studies on-demand.

Now it is time to make this project public for the broader scientific community. Please find it at https://github.com/Emory-HITI/Niffler. This is still an alpha release. But we are working around the clock to make it usable by anyone in the universe of radiology!

 For more details and citation, please follow our pre-print: 

Kathiravelu, Pradeeban, Ashish Sharma, Saptarshi Purkayastha, Priyanshu Sinha, Alexandre Cadrin-Chenevert, Imon Banerjee, and Judy Wawira Gichoya. Developing and Deploying Machine Learning Pipelines against Real-Time Image Streams from the PACS. arXiv preprint arXiv:2004.07965 (2020).

We will keep you updated! Please feel free to send me if you have any questions or suggestions for improvement. I am also up for collaborations, as always! :)