Harnessing AI for STI Care: Image Recognition and Chatbot Innovation

Join us for an interactive workshop exploring cutting-edge applications of artificial intelligence in sexual health. This session will showcase how AI-powered image recognition is revolutionising the diagnosis of skin and mucosal lesions, and demonstrate the construction of a large language model–driven chatbot designed to provide accessible, expert-level STI counselling. Attendees will gain hands-on insights into the development, deployment, and clinical integration of these technologies, with a focus on improving care delivery across diverse global settings.

Background

Medical Artificial Intelligence (AI) is rapidly reshaping the landscape of infectious disease prevention and care, offering innovative tools for risk prediction, decision support, and diagnostic enhancement. This practical one-day workshop will introduce the fundamentals of medical AI with a focus on sexually transmitted infections (STIs). Participants will explore how radiomics techniques can be used to develop image-based screening tools for STIs, such as syphilis, gonorrhoea, and chlamydia, for early detection and build large language model–based chatbots for delivering personalised STI education and triage. Designed for beginners, the workshop assumes no prior knowledge of AI and combines foundational theory with hands-on practice. Attendees will gain step-by-step experience in setting up investigative projects based on simple AI models using text and image datasets, emphasising real-world clinical applications.

Workshop Instructors

Professor Lei Zhang is a health economist and medical AI expert who leads the development of AI tools for STI prevention and control. He pioneered machine learning models for predicting HIV and STI risk among men who have sex with men (MSM) in Australia, and continues to innovate AI-driven approaches to improve access and accuracy in sexual health services.

Jiajun Sun is a PhD candidate at Monash University and an expert in medical AI and image recognition. His research applies advanced radiomics techniques to identify and classify STI-related lesions, with applications extending into dermatology. His work demonstrates the power of AI to support clinical diagnosis through high-resolution image analysis.

Location: Plaza Heisei, 2-chōme-2-1 Aomi, Koto City, Tokyo 135-0064, Japan

Date: June 11, 2025

Time: 9 am to 3 pm

Registration link:https://forms.gle/41R5NaTJhBPVmAYFA (Registration will be closed on May 11, 2025)

Lunch and coffee will be provided to all attendees.

Workshop Agenda

Style: ~30 participants, grouped into 3–4 people per group (9–10 groups)
Platform: Web-based application with integrated Python coding environment and analysis tools; no installation required
Disclaimer: This workshop includes clinical content, including images of sexual anatomy and STI-related lesions, which may be sensitive or graphic

1. Overview (30 minutes)
   Introduction: Medical AI and its transformative role in infectious disease control and diagnosis

• Background: Rapid advances in medical AI and its application in STI surveillance, diagnosis, and decision support
• Demonstration: End-to-end workflow of AI-assisted diagnostic tools used for classifying STI lesions in clinical settings
• Demonstration: Introduction to conversational AI (chatbot) platforms and how they support sexual health consultations

2. AI-assisted Diagnosis of STI Lesions using Radiomics (1 hour 45 minutes)

Theory (45 minutes)

• Concept of radiomics: How quantitative features are extracted from medical images for classification and diagnosis
• Data acquisition: Clinical and imaging data collection processes in sexual health, with discussion of metadata and ethical concerns
• Key features and indicators: Explanation of imaging-derived metrics (e.g., texture, shape, intensity) and clinical annotations
• Radiomics modelling: Overview of algorithms used, including deep learning, random forests, gradient boosting, and ensemble techniques
• Model performance metrics: Interpreting AUC, sensitivity, specificity, PPV, NPV, accuracy, F1 score, and precision for diagnostic evaluation

Practical (60 minutes)

• Dataset: 500 clinically annotated images of STI lesions across multiple anatomical sites
• Image pre-processing: Steps for cleaning, cropping, and normalising clinical image data before analysis
• Feature extraction and modelling: Build and test basic radiomics pipelines to classify warts vs. non-warts
• Model evaluation: Generate and interpret output metrics, including confusion matrices and ROC curves

3. Creation of a Simple Medical Chatbot (1 hour 45 minutes)

Theory (45 minutes)

• Chatbot foundations: Comparison between rule-based and machine-learning-powered chatbot systems used in health contexts
• Natural language processing (NLP): How NLP enables chatbots to process, interpret, and respond to clinical queries
• Chatbot use cases: Examples of chatbot deployment in STI triage, symptom checking, and referral pathways
• Design workflow: Key stages from chatbot script planning, intent mapping, user response training, to backend integration

Practical (60 minutes)

• Script design: Define symptom-related input patterns and clinically appropriate output responses
• Implementation: Use open-source platforms to create a functioning chatbot interface with basic clinical logic
• Testing and refinement: Simulate user interactions, check logic flow, and adjust based on misclassification or ambiguity

4. Ethical and Regulatory Considerations in Medical AI (30 minutes)

• Clinician roles: Will AI supplement or replace clinicians? Discussing the balance of automation and expertise
• Responsibility and harm: If AI misdiagnoses, who is liable — the clinician, developer, or health system?
• TGA and global regulations: Key approval pathways and standards for AI diagnostic tools in Australia and internationally
• Privacy and data ethics: Navigating consent, de-identification, and trust in AI used for sensitive health issues

5. Multi-country Collaboration via Data Consortium (30 minutes)

• Cross-context adaptability: How AI diagnostic tools trained in one setting can be validated and refined in another
• Consortium structure: Building a collaborative STI image data consortium across countries (e.g., Australia, China, Brazil, South Africa)
• Shared learning: Facilitating model improvement and validation through pooled datasets and diverse population data
• Transfer learning: Using shared features to adapt AI models across regions with different disease profiles and imaging styles
• Ethics and benefit-sharing: Ensuring equitable access, intellectual property agreements, and capacity building for all participating partners