top of page

Empowering AI Innovations Together //

Cross-Adaptation

Updated: Oct 11




Project Overview

Our team was faced with a challange to improve AI-powered COVID-19 detection system. The goal was to improve accuracy and deployability across diverse patient populations globally.


The Challenge

The existing AI model for detecting SARS-CoV-2 from routine blood tests showed promise but faced significant hurdles:

  • Inconsistent accuracy across different regions

  • Extensive retraining required for each new deployment location

  • Slow time-to-deployment, hampering rapid response capabilities


Our Solution: Implementing Cross-Adaptation

We proposed and implemented an innovative "cross-adaptation" method to overcome these challenges:

  1. Universal Adaptability: Enabling the system to adjust to data from various sources without extensive retraining.

  2. Improved Accuracy: Achieving up to a 10 percentage point increase in F1 score on average across datasets.

  3. Faster Deployment: Eliminating the need for labeled data from each new target population


Implementation Process

  1. Data Integration: Harmonized data from 9 datasets spanning multiple countries.

  2. Algorithm Development: Created a cross-adaptation algorithm compatible with various classification methods.

  3. Rigorous Testing: Validated the system using 5 machine learning algorithms and 3 domain adaptation methods.

  4. Performance Optimization: Iteratively refined to maximize accuracy and adaptability.


Results and Impact

The implementation of our cross-adaptation solution led to significant improvements:

  • Accuracy Boost: F1 scores improved from an average of 62% to 72% across different classification models.

  • Deployment Efficiency: Time-to-deployment in new locations decreased by 60%.

  • Cost Reduction: Significantly reduced need for extensive local data collection and model retraining.

  • Enhanced Scalability: Enabled rapid expansion to new markets with minimal additional investment.


Project Outcomes

The success of this project has had far-reaching implications:

  • Improved Patient Care: Faster, more accurate COVID-19 detection across diverse populations.

  • Competitive Advantage: The AI system can now be deployed globally with greater speed and accuracy.

  • Cost Savings: Reduced need for localized data collection and model adaptation.

  • Market Expansion: Facilitated entry into new healthcare markets previously challenging to serve.


Comments


bottom of page