Artificial intelligence is changing how we detect diseases. One exciting area is the use of deep learning to identify sexually transmitted infections (STIs) from skin lesion images.
A recent review in The Lancet Digital Health (July 2025, Vol 7) looked at how well these AI models work. The results are promising, but there are still challenges to overcome before they can be used widely in clinics.
Why Early Detection of STIs Matters
STIs are a major global health issue. In 2019, there were over 770 million new cases. Catching infections early helps people get treatment faster. It also stops the spread and reduces long-term health problems. However, in many low- and middle-income countries, lab tests aren’t always available. That’s why new tools like AI are so important.
Deep learning can help by analyzing skin lesion images. It may detect infections like mpox, scabies, herpes, syphilis, and molluscum contagiosum—even before lab tests confirm them.
How Deep Learning Is Being Used to Detect STIs
The review looked at 101 studies published between 2010 and 2023. A deeper analysis of 55 studies showed strong results:
- Mpox detection: 97% sensitivity, 99% specificity
- Scabies detection: 95% sensitivity, 97% specificity
Most studies used CNN models like ResNet and VGGNet. These models were trained on public datasets such as MSLD, MSID, and Monkeypox2022. However, only a few studies used newer models or included extra data like age, sex, or skin tone. This extra data can help improve accuracy, especially for diverse populations.
Deep Learning Models: Key Quality Challenges in STI Detection
Even though the results look good, many studies had problems that could affect how well the models work in real life:
- 85% of studies used public datasets lacking diversity in skin tone and body location.
- Only 2% employed prospective data collection.
- 98% failed to report race and ethnicity.
- 93% did not describe labelling methods.
- Only 6% shared source code for public evaluation.
Because of these gaps, it’s hard to know if the models will work well in clinics, where conditions vary a lot.
Improving Trust and Transparency in Deep Learning for STI Diagnosis
Trust is key when using AI in healthcare. Some studies used explainable AI (XAI) tools like Class Activation Maps (CAM) and LIME. These tools show which parts of an image the model focused on. But without proper testing, they might confuse rather than help doctors.
A Roadmap for Responsible AI in STI Detection
To make these tools more useful, researchers suggest a clear plan. Here’s what they recommend:
- Create diverse and standardized datasets.
- Test models in real-world settings.
- Include metadata like age and skin tone.
- Use XAI tools to improve transparency.
- Develop medical guidelines for AI use in STI screening.
Also, models should be small and fast enough to run on mobile devices. Privacy tools are important too, especially when dealing with sensitive images.
Source: The Lancet Digital Health, Vol 7, July 2025. DOI: 10.1016/j.landig.2025.100894
