Helfie STI AI
Research Paper Summary

Overview

Helfie.AI is backed by science. 

​

At Helfie we work closely with reputable and published research partners to design and develop the health AIs that underpin our technology.

​

Here’s  a summary of the scientific study titled “Clinical features-based machine learning models to separate sexually transmitted infections from other skin diagnoses”  by Professor Christopher Fairley from our research partners Monash university & Alfred Health.
 

The STI AI model, developed through a collaboration between Helfie’s research partners Monash University and Alfred Health, is a cutting-edge digital tool designed to differentiate sexually transmitted infections (STIs) from other skin-related diagnoses based on clinical features alone. This AI-driven model analyzes a range of patient-reported symptoms and clinician-observed factors—such as lesion type, number, and specific genitourinary symptoms (including dysuria and anorectal pain)—to help quickly assess whether a condition is likely an STI or another type of skin issue. The tool’s goal is to support healthcare providers by enabling faster, more accurate STI triaging, especially beneficial for overburdened health services where demand exceeds capacity.

The development study, conducted at Melbourne Sexual Health Centre, utilized data from 1,315 patient records, focusing on a broad array of symptoms commonly associated with STIs and other skin conditions. To maximize the model's accuracy, researchers tested 16 machine learning algorithms, selecting the CatBoost model as the top performer. CatBoost achieved a notable area under the curve (AUC) of 0.912, an accuracy of 83.7%, sensitivity of 77.6%, and specificity of 83.1%. This means the model could effectively identify STI cases, while also minimizing false positives, making it a valuable tool for initial STI screening in clinical settings where rapid assessments are essential.

Key indicators for the model’s success included both visible dermatological symptoms (like lesions) and other clinical features, highlighting that certain skin lesion types, such as ulcers and sores, were more strongly associated with STIs, while conditions like rashes and white patches often pointed to non-STI diagnoses. Additionally, symptoms such as lesion pain and duration, along with demographic factors (like age and gender), contributed significantly to model predictions. This data-driven approach offers a novel solution for STI differentiation without the need for immediate laboratory testing, streamlining the screening process and conserving healthcare resources.

While this model has shown strong performance with clinical features alone, the study’s authors suggest that incorporating clinical images of skin conditions could improve diagnostic accuracy even further. This additional layer could make the tool especially reliable for healthcare environments that rely on remote diagnosis or telemedicine services, where quick, non-invasive assessments are crucial. The research highlights the potential for digital health innovations to provide accessible, high-quality STI diagnostics, enhancing the ability to manage STI spread efficiently.

This model represents a significant step toward accessible, AI-powered health screening tools like Helfie,  tailored to meet the growing demand in sexual health services.

To explore the research and in-depth findings supporting this breakthrough, you can read the full article here.