Abstract B026: Artificial Intelligence Apps for Medical Image Analysis using pyCERR and Cancer Genomics Cloud
Bibliographic record
Abstract
Abstract Introduction: This work introduces cloud-based Apps for Artificial Intelligence analyses of medical images using NCI-funded Cancer Genomics Cloud (CGC) and Python-based Computational Environment for Radiological Research (pyCERR) platform. It fills the need for a ready to use software for applying AI models to medical images without requiring software and hardware resource setup from the end user. Methods: The components of the framework consist of: (i) AI Apps: Analysis workflows involving AI models are packaged as “apps” that are launched in Docker environments and can be configured to use hardware resources from commercial cloud providers like AWS, GCP and Azure. Apps are defined in Common Workflow Language (CWL) and multiple CWL tasks can be chained together in a workflow definition. These Apps are initiated as tasks in the cloud within the CGC - Seven Bridges web portal or externally from users’ local hardware using their API via Python or command line. These apps along with the required data are organized into projects on CGC; and shared with other users on the platform for their use. (ii) Data I/O: The framework allows datasets to be downloaded from public repositories such as TCIA, IDC, Zenodo as well as XNAT instances by using APIs to download images. The CGC-API provides upload and download from users’ local hardware to CGC, allowing easy transfer of user data for AI inference. Trained AI model weights are hosted in access-controlled Box software at our institution and made available to Apps on-demand. Additionally, users can store and programmatically access data on cloud storage provided by GCP and AWS. (iii) Analysis software: GNU-GPL copyright pyCERR, a Python-based computational platform enables researchers to organize, access, and transform metadata from high dimensional, multi-modal datasets. pyCERR provides an extensible data structure for metadata from commonly used Radiological imaging file formats and Napari-based Viewer to allow for multi-modal visualization and programmatic access to graphical user interface objects. Analysis modules in pyCERR provide image segmentation, radiomics, Dynamic Contrast Enhanced MRI features, radiotherapy dose-volume histogram-based features, and normal tissue complication and tumor control models for radiotherapy. Image processing utilities are provided to help train and infer convolutional neural network-based models for image segmentation, registration and transformation. Results: The framework provides access to AI models clinically used at our institution for research, non-commercial use. These include segmentation inference for various treatment sites of Organs at Risk in radiotherapy treatment planning, lung tumor nodules and deformable image registration for abdominal MRI in the form of Apps hosted in a CGC project. Conclusion: In summary, the presented framework facilitates reproducible deployment of radiological image analyses by pre-configuring components of analyses including hardware resources and computation. Citation Format: Aditya Apte, Eve LoCastro, Aditi Iyer, Sharif Elguindi, Jue Jiang, Jung Hun Oh, Harini Veeraraghavan, Amita Shukla-Dave, Joseph O. Deasy. Artificial Intelligence Apps for Medical Image Analysis using pyCERR and Cancer Genomics Cloud [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr B026.
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 | 0.002 | 0.005 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.003 | 0.004 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.025 | 0.018 |
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".