Related Experiment Video
Updated: Aug 12, 2025
08:58
Artificial Intelligence Approaches to Assessing Primary Cilia
Published on: May 1, 2021
3.6K
MIDRC CRP10 AI interface-an integrated tool for exploring, testing and visualization of AI models
Naveena Gorre1, Eduardo Carranza1, Jordan Fuhrman2
1Department of Machine Learning, Moffitt Cancer Center, Tampa, FL, United States of America.
Physics in Medicine and Biology
|January 30, 2023
Summary
This study introduces the CRP10 AI Application Interface (CRP10AII), a user-friendly tool for developing and interpreting machine learning models in clinical applications. CRP10AII simplifies AI model building and visualization for researchers and clinicians.
Area of Science:
- Medical Artificial Intelligence
- Machine Learning in Healthcare
- Radiology and Oncology AI
Background:
- Developing clinical Machine Learning (ML) models is complex, requiring expertise and facing challenges with incompatible frameworks and data preparation.
- Diagnostic radiology and oncology applications present unique difficulties due to heterogeneous data and specialized task requirements.
- Existing tools often lack user-friendliness and comprehensive interpretability features for AI model development.
Purpose of the Study:
- To provide a modular, flexible, and user-friendly software tool for clinicians, researchers, and early AI developers.
- To enable effective exploration, training, and testing of AI algorithms, with a focus on model interpretability and explainability.
- To allow visualization of performance and interpretation of predictions across neural network layers for deep learning algorithms.
Main Methods:
- Development of the CRP10 AI Application Interface (CRP10AII) using the Django web framework in Python.
- Integration with data commons platforms like Gen3 for data acquisition and management.
- Implementation of features for querying data, training models (from scratch or transfer learning), evaluating performance, and interpreting predictions using methods like SHAPLEY and LIME.
Main Results:
- CRP10AII facilitates data acquisition, model training (including transfer learning with models like VGGNet, AlexNet, BERT), and performance assessment (e.g., ROC curves).
- The tool enables interpretation of AI model predictions through SHAPLEY and LIME values.
- Visualization of model learning is achieved via heatmaps and activation maps of individual neural network layers.
Conclusions:
- CRP10AII offers an end-to-end solution for users to pre-process, build, train, visualize, and explore AI models, significantly reducing the time and expertise needed.
- The open-source nature of CRP10AII encourages community development and adaptation based on user feedback.
- This tool democratizes AI development in clinical applications, particularly for users with less experience in machine learning.
