RAIVEN: A novel framework for translation of AI tools to the radiology environment
Bibliographic record
Abstract
1183 Introduction: Artificial intelligence (AI) is finding an increasing number of applications in nuclear medicine and radiology. Deep learning models have been used to improve image quality, automate image segmentation, and aid in disease classification. Models that combine learned features, radiomic features, and clinical information can be used to predict disease progression, treatment outcome and survival. However, translation of such models/algorithms into clinical trials or standard of care applications is often nontrivial due to the lack of an appropriate framework for model deployment. The aim of this project is to develop such a framework that satisfies the following design specifications: 1) Allows easy combination of different algorithms to create pipelines with multistep workflows, 2) Works with existing workstations available to physicians/researchers, 3) Allows the deployment of models developed in any programming language without the need for cumbersome and time-consuming configuration, 4) Is easy to deploy and use. Our framework, RAIVEN (Radiology AI Virtual ENvironment), represents our vision of a new radiology environment that incorporates AI tools to better treat and diagnose disease. Methods: The application encompasses a central API service and five auxiliary modules, including a database, worker daemons, a messaging queue, a frontend, and a DICOM service. The API, developed using the asynchronous Python framework FastAPI, is the main component of the application. It controls all communication between the five auxiliary components. The DICOM service, implemented using the Python package pynetdicom, governs both the input and output of DICOM images. RAIVEN conforms to all DICOM networking standards; it can be integrated seamlessly with all DICOM enabled software and equipment present in nuclear medicine and radiology departments. Our framework allows users to build processing pipelines using a visual interface by connecting different algorithms developed by researchers. These algorithms are added to RAIVEN as Docker containers. With Docker as the underlying mechanism for our application, tools are easy to update, language agnostic, and maintained in separate virtual environments. The client-side, developed in VueJs, provides an accessible user interface running on all modern web browsers. Conforming to material design principles, the interface delivers a simple, yet informative user experience. The web application is also mobile-friendly, allowing users to easily create, edit, and run pipelines from a variety of devices. Results: We present a framework to enable imaging pipeline creation. Users can upload containerized algorithms, use drag and drop to connect containers, move received DICOM images through pipelines, and download/export resulting DICOM images or other generated filetypes. Our application differs from existing solutions published by providing a unique interface for users to visually connect tools to build medical image processing pipelines. We have deployed the application as a webservice hosted internally at our institution. RAIVEN has been tested with image processing workflows including algorithms for image anonymization, and simple image processing algorithms such as down sampling and pixel manipulation of PET/CT and SPECT/CT images. Researchers can easily add more complex image processing methods. The source code of the application is publicly available at github.com/qurit/raiven. Conclusions: RAIVEN is an open-source framework that aims to facilitate translation of AI research tools into the clinical environment. The ease of constructing a pipeline encourages users to create many workflows to test their developed algorithms, facilitating new discoveries and quicker diagnosis. We envision RAIVEN, given its various aforementioned capabilities, to speed up the clinical translation of image processing algorithms at our functional imaging center and at other centres around the world.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.007 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.001 | 0.004 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.005 | 0.004 |
| Open science | 0.007 | 0.011 |
| Research integrity | 0.002 | 0.005 |
| Insufficient payload (model declined to judge) | 0.020 | 0.010 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".