Hertility: Bayesian AI Transforms Women's Health

An Innovative Approach to Women's Health

In the often-overlooked field of women's health, Hertility stands out with its innovative approach. During a discussion on Just Now Possible, Teresa Torres explored the advancements of this company with Tulsi Patel, Lorna Brightmore, and Jack Pickard, key figures at Hertility. Based in the UK and Ireland, the company offers a unique combination of online health assessments, at-home hormonal testing, and clinical reports. These tools aim to diagnose conditions ranging from menstrual disorders to menopause.

A Rich and Diverse Database

Hertility relies on an impressive database accumulated over seven years. This collection of data includes symptoms, blood test results, and pelvic ultrasounds from over one million women. This information is essential for the development of two major AI products: Gyn.AI and a scan automation system. Gyn.AI uses a Bayesian network to provide probability-based diagnostics, while the automation system analyzes ultrasound images to measure key parameters such as follicle count and ovarian volume.

How Gyn.AI Works

Gyn.AI is distinguished by its use of a Bayesian network. Unlike traditional systems that provide binary diagnostics, Gyn.AI offers probabilistic diagnostics. This approach allows clinicians to better understand the nuances of the results, thereby enhancing their confidence in AI tools and facilitating faster and more accurate triage.

Transparency and Trust

Transparency is a central pillar of Hertility's strategy. By showing clinicians not only the diagnosis but also the reasoning behind it, the company builds trust in its tools. This transparency is crucial for speeding up the triage process and improving care efficiency.

Combating Automation Bias

To avoid automation bias, Hertility employs validation sets and ensures that independent reviews are conducted. This approach guarantees that AI models remain objective and reliable, thereby minimizing the risk of errors in diagnostics.

Scan Automation and Increased Accuracy

Hertility's scan automation pipeline offers superior accuracy compared to traditional manual methods. By classifying ultrasound images and measuring critical parameters, this system enhances the precision of medical diagnostics, providing more reliable results for patients.

The Agentic Loop: An Essential Check

Hertility has implemented an agentic loop to verify the clinical letters generated by AI. This cross-checking with patient data helps identify potential errors before a clinician reviews the results, ensuring the reliability of the information provided.

Infrastructure Challenges and Data Security

The secure transfer of DICOM ultrasound images from third-party providers to Hertility's systems is a major challenge. The company ensures that these transfers are conducted securely to protect patient data confidentiality.

Managing Sensitive Data

Hertility takes rigorous measures to protect personally identifiable information (PII) and protected health information (PHI). By employing pseudonymization and data minimization, and running models on AWS Bedrock, the company ensures that sensitive data is handled responsibly.

Regulation as a Design Driver

Rather than viewing regulation as a hurdle, Hertility integrates it from the outset as a design constraint. This proactive approach enables the creation of AI products that are not only compliant but also scalable and robust, meeting the strict requirements of the healthcare sector.

Conclusion: Lessons to Learn

Hertility's AI solutions, by offering probabilistic and transparent diagnostics, enhance clinician trust. Measures to protect against automation bias are as crucial as the development of the model itself. Finally, data minimization and a solid internal infrastructure allow Hertility to handle sensitive health data responsibly, while making its AI products more defensible and scalable.