AI-Driven Healthcare Solution for Symptom Assessment & Test Analysis
During the COVID-19 pandemic, there was a surge in demand for COVID testing, and our developed application proved invaluable in expediting turnaround times and prioritizing the testing order of diagnostic samples from patients.
Overview
Our client experienced a significant surge in demand for testing services each day during the COVID-19 pandemic and they recognized an opportunity to significantly enhance operational efficiency and save costs. However, they lacked the technological means to scale-up the process.
To accomplish this, we developed an ML model, integrated with the existing application for predictive analysis, facilitating easier patient sample screening and lowering overall operational expenses.
About the Project
We performed the role of a technology thought leadership partner through our CTO-as-a-service offering to enable AI-based smart pooling for our client to streamline operations in diagnostics lab, specifically in the testing process.
Initially, our client conducted individual testing for each sample and pooled a small number of samples to minimize testing errors. Scaling up the pooling process could enhance operational efficiency and significantly reduce costs. However, it also posed the risk of longer turnaround times and increased costs if not executed correctly.
To mitigate the risk of pooling incorrect samples during testing, we have developed a ML model possessing 88% accuracy. This model performs predictive analysis to streamline testing processes and lower associated costs.
Our Experience
Our ML solution was designed by drawing on insights from continuous experimentation and coordination with the SDI Labs Team conducted over the years. Some of the impacts of the ML solution for our client are:
• Reduce turnaround time
• Increase efficiency
• Enhance reliability and accuracy of results
• Seamlessly integrate with existing workflows
country
United States of America
industry
Healthcare
timeline
2020 - Ongoing
3+
Years in Partnership
200,000+
Patients Health Risk Analysed
3+
AI IPs Developed
88%
Accuracy for True Positive Identification
ML System Workflow
Developed Solutions
We developed an ML model which was integrated with the existing application, to provide test recommendations based on symptom assessment and the patient’s risk profile, facilitating smart pooling of patient samples.
ML Model Architecture
The model was built to predict the probability of a person testing true positive or true negative for COVID-19. Weightage was assigned to each question in the health intake form, reflecting its impact on the final output and backed by science. Furthermore, this model integrates explainable AI, offering insights into the reasoning behind its output, a critical feature for healthcare-focused AI models.
Training of ML Model
This ML model was trained using the health intake forms and their corresponding test results to enable our client to effectively pool the most probable true negative result bearing samples together. Over 35000 health intake forms were used for training the model to an accuracy of 88% of predicting true COVID-19 positive patients.
AI-based Smart Pooling
Pooling samples identified as true negatives by the model's output can optimize testing processes, leading to reduced operational costs compared to traditional workflow and can aid in streamlining operations. Additionally, this approach allows for scalability in the pooling process.
