Nillion Health

Nillion Health is a program that demonstrates the power of blind computation in a healthcare setting. Making use of Nillion's multiparty computation, the program provides a breast cancer image classification test performed over multiple health providers.

Problem:

The article "Sharing Is Caring-Data Sharing Initiative in Healthcare"[1] highlights a concern in regards to who gets access to all the health data. High valued data is expected to further grow in value due to advances in AI/ML and other emerging technologies. Do you know if your health provider can provide you the best care when you need it?

Objective:

Make use of modern cryptography to prevent barriers in the healthcare industry and ensure the world's most important data type, medical imaging, can be shared in a way that is fair and secure.

Demo

The diagram illustrates the program's demo. There are two primary sections, a test computed over a full dataset and a test computed on Nillion's protocol on combining multiple datasets.

Full Dataset Test

• 550+ image data instances
• 30 parameters and 1 target value (Diagnosis)
• 80/20 Split, (80) Training data, (20) Testing
• 3 Figures of Plot Distributions - Full, Small Subset, Large Subset
• Simple classification model use for computing thetas and test predictions
• Single randomly selected test instance used, not a full test evaluation. Do not use program for predictions. Purpose of testing is to compare calculated values.

Multi Party Nillion Test

• Small Subset (25%) and Large Subset (75%)
• 30 parameters and 1 target value (Diagnosis)
• Weighted average method used when combining thetas from subsets
• A scaling factor was applied to satisfy integer requirements in Nillion's Nada program

About the program

✅ Secure Multi-Party Computation (Blind computation)

• Nillion - Nillion is a secure computation network that decentralizes trust for high value data in the same way that blockchains decentralized transactions.

✅ Diagnostic Imaging Cancer Data

• UC Irvine Machine Learning Repository - The dataset used in the program is related to diagnostic imaging data focused on breast cancer classification. As included in the site's information, the features are computed from a digitized image of a fine needle aspirate (FNA) of a breast mass. It describes characteristics of the cell nuclei present in the image.

✅ Files

• Main Python Program
• Nillion Nada Program
• Dataset

✅ Setup

Requirements

• Python/Python3
• Terminal

Clone repo and change into nillion-python-starter directory

$ cd nillion-python-starter

Create a python virtual environment and activate

$ python3 -m venv .venv/

source .venv/bin/activate

Install dependencies

pip install -r requirements.txt

Initialize Nillion environment and compile Nada program

./bootstrap-local-environment.sh

./compile_programs.sh

Change to healthcare_imaging_compute directory and run program

cd healthcare_imaging_compute

python3 healthcare_imaging_compute.py

✅ References

[1] Hulsen T. Sharing Is Caring-Data Sharing Initiatives in Healthcare. Int J Environ Res Public Health. 2020 Apr 27;17(9):3046. doi: 10.3390/ijerph17093046. PMID: 32349396; PMCID: PMC7246891.

✨ Project Member 👨‍💻

Created for ETHGlobal Scaling Ethereum 2024 by:

• Chris Salvador