MIDRC Helper AI and Pixel DeID:
Enhancing Medical Imaging Analysis with AI

The MIDRC team is committed to advancing the field of medical imaging through innovative technology and thorough data preparation. Recently, we've developed two AI models designed to enhance the accuracy and privacy of imaging data: MIDRC Helper AI and Pixel DeID.

The MIDRC Helper AI model improves the accuracy of imaging metadata by annotating data with correct anatomical information, orientation, and contrast. For example, Helper AI can effectively address issues like mislabeled studies, which can significantly improve research outcomes. By deploying this model, we can enhance both the reliability and usability of medical images.

Pixel DeID addresses the challenge of de-identifying pixel-level data in medical images. This model automatically detects sensitive information burned into the images. The detected pixel regions can then be automatically redacted. For example, Pixel DeID + Redaction can automatically detect protected health information (PHI) like patient names burned into the image and redact them. This automated tool not only saves time, but also minimizes the risk of human error. It ensures that medical images are shared for research purposes without jeopardizing patient privacy.

These tools align with MIDRC’s mission to provide high-quality, de-identified medical imaging data for research. As we continue to refine these models, our focus will expand to include a wider variety of imaging modalities and conditions. For in-depth demonstrations and to see these models in action, please check out the video on our YouTube channel.

MIDRC Diversity Calculator:
Assessing the Representativeness of Biomedical Data

The MIDRC Diversity Calculator enables researchers to assess the representativeness of biomedical data. This open-source tool uses the Jensen-Shannon distance (JSD) measure to evaluate demographic representativeness, ensuring that biomedical data reflects diverse populations.

Key Features of the Diversity Calculator:

1. Jensen-Shannon Distance (JSD) Calculation: Assesses how closely data aligns with target population demographics.
2. Comparative Analysis: Compares different datasets to highlight diversity gaps, a crucial feature for ensuring equity in biomedical research.
3. Historical Data Monitoring: Tracks changes in data representativeness over time, enabling researchers to monitor changes and track progress toward more inclusive datasets.
4. Biomedical Focus: Specifically tailored for analyzing biomedical data (both imaging and non-imaging), ensuring that comparisons are relevant and meaningful within the context of health research. 
5. Generalizability: Adaptable to various datasets and demographics, with easy data input via Excel.

1. Expanding AI-Ready Data Commons: Ensuring representative data for a well-curated, diverse repository following FAIR principles (Findable, Accessible, Interoperable, and Re-usable).
2. Fostering Machine Intelligence Research: Facilitating the development of trustworthy algorithms for disease-related AI applications by providing diverse datasets.
3. Supporting Data Management and Sharing: Enhancing data quality and applicability, in line with the NIH Data Management and Sharing Policy.

Conclusion
The MIDRC Diversity Calculator is crucial for promoting inclusive and representative biomedical AI/ML research. By ensuring data diversity, MIDRC advances research and health equity.

For more information and to access the Diversity Calculator, visit MIDRC's website. The software is also available on the MIDRC GitHub page, and you can find a link to our algorithms page here. 

• CheXpert Plus Dataset
• arXiv Paper
• De-ID Algorithm Paper
• Hugging Face De-ID Algorithm
• Tweet Thread