Transforming Medical Clinical Practice with AI

A few years ago, during a family trip to a remote area in Brazil, my daughter suddenly began scratching her leg frantically.

Alarmed, I told her, “Come here, let me see what you’ve got”.

I discovered a huge, inflamed cross on her leg as if some big, strange insect had bitten her. I gave her an antihistamine and applied a cream.

Moments later, she complained: “Mom, I also feel something in my throat, like it’s hard to swallow.”

Panic set in! Difficulty swallowing could mean something serious. We ran out, and after a two-hour drive, we finally reached the hospital.

The doctor who examined her looked horrified, left the room without a word, and returned with two more doctors. After a hurried consultation, they administered an antihistamine injection and prescribed a cream but offered little explanation.

The next day, as soon as it dawned, I went to buy the prescribed cream. The pharmacist, also a biologist, asked if I knew what it was for. I had no idea.
It is Larva Migrans, he said: A parasitic infection that burrows into the skin, creating map-like patterns. Hence the name Geographic Worms.

With this information — thanks kind Brazilian biologist! — I researched on the internet and discovered that just using the cream wouldn’t be enough. She also needed to take an oral antiparasitic.

The next day, back in my country, we got off the plane and took her directly to the emergency room. The doctor examined her leg and said:

“I don’t know what she has.”

A bit annoyed, I replied, “I know what she has!” and explained the diagnosis to her. I expected her to look up this new information on her computer to understand the condition and treatment options, but she didn’t. She said, “I’m going to prescribe a generic antiparasitic, Z.”

In my research, I concluded the antiparasitic had to be X or Y. So, I got Y while waiting for a dermatologist appointment.
The dermatologist confirmed the diagnosis, saying Y was fine but X would be better, not Z.

After three days, my daughter was cured!

AI in action

Recently, while preparing for an AI presentation, I remembered this incident (how could I forget!) and decided to test the scenario with AI.
I wanted to test it by uploading only a photo of her leg, but I couldn’t find any as I had deleted them.

So, I typed the following prompt in an AI chat:

In two seconds, AI gave me the exact diagnosis, its descriptions, and the exact treatment! (With X or Y medication, not Z :) ).
It also gave me other options, with less probability but worth considering.

Why am I sharing this story?

I can understand that the emergency doctor didn’t know about a condition common in Brazil but rare here. What upset me was that she didn’t turn on her computer to look it up!

Today, she wouldn’t even need to turn on a computer. She could take a picture of my daughter’s leg, upload it, and get a possible diagnosis and treatment. In my case, we could have saved time, worry, and the cost of a bad treatment.

I grew up with doctor parents, so I know about the challenges doctors face. Today, with my sisters and niece also being doctors, I hear the same problems from 30 years ago, along with new ones caused by poor and complicated medical software.

I am an engineer, so let’s explore the potential of AI to help doctors.

The Problems

Imagine you are a clinical doctor who attends to patients, so probably:

1. You have limited time to attend to each patient.
   In many healthcare institutions, appointments must last no more than 10–15 minutes. Most doctors agree it is difficult to provide good care with these restrictions.
2. You need to quickly understand the patient’s current situation and history when they arrive.
3. There’s a lot of information that doctors have but can’t consider during the consultation due to its length and complexity. The patient’s medical history, for example, is underutilized.
   (This reminds me of when my sister, in her first clinical job, took home patient records to study the day before so she could provide good care in just 10 minutes)
4. It can be challenging to recall and apply all of your vast medical knowledge during patient consultations. More needed if you don’t have much experience yet.
5. New research, treatments, and medications are constantly emerging. You must stay up-to-date to provide the best care.
6. You need to update the clinical history with the examination results, diagnosis, and treatment, all in a structured format, which is often a tough task.
   (My other sister mentioned that it often takes more time and effort to enter this data into structured forms than to diagnose and treat the patient.)

Given all these conditions during an appointment, you must make important decisions that affect people’s lives.

AI technology provides us with a huge opportunity to solve all these problems and others, making much better use of the scarce 10 minutes of clinical consultations.

Let’s imagine a solution with this objective.

The Solution

The solution could be something like this:

1. Once the patient enters your office, the app identifies them by their appointment.
2. You will find a summary of the patient’s clinical history on the screen, enabling a quick understanding of their situation and the option to delve into more detail.
3. You can enter the clinical examination results through voice, picture, or text, as you want, to simplify the record-keeping process.
4. You will have an AI assistant to ask in natural language about anything. Questions can include:

• Anything about the patient’s medical history.
• Related medical knowledge, the latest research, and information about similar cases.
• Medication options, substitutes, and availability.
• Disease prognoses and medical statistics.
• And much more

Think of this AI assistant as an expert doctor familiar with your patient’s clinical history, with quick access to vast amounts of information to discuss patient conditions.

It can also offer new perspectives and provide insights on diagnoses, treatments, medications, substitutes, medical statistics, and more.

5. Finally, you can record your conclusions and treatment: by voice, picture, text, whichever is most practical for each doctor.

Unlocking AI’s Potential

This is an example of the innovative AI-driven applications we can build today.

AI can revolutionize medical practice by providing accurate, quick, and comprehensive assistance to doctors, considering details of the patient’s medical history and updated medical knowledge, and simplifying data entry.

While AI cannot replace medical judgment, it can enhance practice, leading to better patient outcomes, reduced stress for medical professionals, lower healthcare costs, and improved healthcare services.

Technical Solution

If you’re a technical person interested in implementing this type of solution, I will be publishing an overview of the concepts and techniques on how to build this solution using LLMs, RAG techniques, vector databases, and embeddings in the coming days.

If you’re not familiar with these techniques, you can learn about them and get inspired to incorporate them into your applications.

AI Series

This article is part of a series on the fundamentals of AI. I will be posting more in-depth explorations of these topics, analyzing various use cases, including some that my team and I have worked on.

• AI: How to Create the New Generation of Applications — published
• Transforming Medical Clinical Practice with AI —this article
• How to Implement an AI Assistant to Improve Medical Practice — coming next

Found this article useful? Follow me (Mayda). I post periodically about App Design, AI, UX, R&D, and Neuroscience, aiming to turn complexity into clarity — first in my mind, and hopefully in yours.