Summary of "1º Treinamento de Inteligência Artificial em Medicina"

Summary of Main Ideas and Concepts

Artificial Intelligence (AI) is transforming the world and medicine.
CREMESP (Regional Medical Council of São Paulo) is pioneering the first AI training dedicated to healthcare professionals.
AI is a tool to support doctors, not replace them.
The training aims to provide practical learning, ethical reflection, and critical discussion on AI's role in medicine.
AI impacts all stages of medicine: prevention, diagnosis, treatment, and monitoring.

AI has been evolving for decades; recent advances include Large Language Models (LLMs) like ChatGPT.
LLMs represent a breakthrough because they process and generate natural language, making AI accessible without programming skills.
AI is already embedded in everyday tools (e.g., Netflix recommendations, Google autocomplete).
The iPhone analogy: AI itself is a platform, but the apps built on it will revolutionize daily life and medicine.

AI aids in diagnostic imaging, clinical decision support, predictive analytics, and robotic surgery.
Examples include:
- Mirai AI predicting breast cancer 5 years earlier than traditional methods.
- AI algorithms highlighting bleeding in CT scans.
- Tools predicting cardiac events based on tomography.
- Augmented reality glasses guiding surgeons.
- AI monitoring voice changes to predict Alzheimer’s.
- AI models forecasting sepsis and transfusion needs.
- AI-guided robotic surgeries, including autonomous gallbladder removal.
AI can analyze vast datasets beyond human capacity, improving diagnosis and treatment planning.
AI tools require patient consent and ethical oversight.

AI models are probabilistic and can produce variable or incorrect answers ("hallucinations").
Black-box nature: difficulty understanding how AI arrives at conclusions.
Dependence on data quality and potential biases.
Regulatory and ethical concerns, including accountability when AI errors occur.
Energy consumption is a major limitation for widespread AI use.
AI updates can unpredictably affect outputs, posing risks in clinical use.
Risk of misinformation and prompt injection attacks (manipulating AI outputs).
Need for human supervision and validation of AI recommendations.

AI regulation is critical to ensure safety, ethics, and transparency.
The EU has developed a comprehensive AI regulatory framework classifying AI systems by risk:
- Minimal risk (e.g., video games)
- Limited risk (e.g., consumer interaction systems)
- High risk (most medical AI applications)
- Unacceptable risk (e.g., social scoring, remote biometric ID)
High-risk AI systems require rigorous risk management, human oversight, and compliance audits.
The EU created a centralized AI supervisory office and national authorities for enforcement.
Regulatory sandboxes allow safe testing of AI systems before market release.
The EU also regulates medical devices using AI, requiring compliance with both AI and medical device regulations.
Data governance is essential; the EU promotes a European Health Data Space for secure, cross-border data sharing to support AI development and healthcare delivery.
Transparency and explainability are mandated to ensure user trust.

Doctors should learn to interact effectively with AI, especially in constructing good prompts.
A four-step prompt methodology:
1. Define the persona (e.g., cardiologist, radiologist)
2. Provide detailed context (patient history, symptoms, exam results)
3. Ask a clear, specific question (diagnosis, treatment options)
4. Request clarifying questions from AI to refine answers
Treat AI as a "co-worker," not just a tool.
Examples of AI use:
- Dermatoscopy image analysis.
- X-ray interpretation.
- Electrocardiogram reading.
- Complex diagnosis (e.g., microscopic polyangiitis).
Use multiple AI tools to cross-verify information.
Study AI daily, even for short periods, focusing on relevant specialties.
Use AI for non-medical tasks initially (e.g., recipes, language learning) to build familiarity.

GPT Chat (latest GPT-5 version) with multiple modes (fast, deep thinking, agent mode).
Specialized AI tools for medical use (Open Evidence, Geni, Voa Health).
AI-powered transcription and medical record summarization.
AI catalogs (e.g., Futuripia) to find tools by specialty.
Tools for creating presentations, images, music, and advertisements.
Google’s Gemini AI and other emerging AI models.
Notebook LM for interactive study and podcast creation from medical content.