How to create fake medical images[Technique Tuesdays]

Don't worry, we won't be doing anything shady with this technique

Dec 20, 2022

To learn more about the newsletter, check our detailed About Page + FAQs

To help me understand you better, please fill out this anonymous, 2-min survey. If you liked this post, make sure you hit the heart icon in this email.

Get a free Weekly Summary of the important updates in AI and Machine Learning here

Recommend this publication to Substack over here

Take the next step by subscribing here

I’m sure you’re very sketched out by the title,

Don’t worry, we’re safe. What I’ll be covering today is a technique that helps you solve some of the biggest problems in Medical Research. Understanding this is key for any of you that might have aspirations of applying your tech skills in the Medical Field.

In this article, I will be sharing with you what the researchers who published “SinGAN-Seg: Synthetic Training Data Generation for Medical Image Segmentation”, a particularly promising approach, did. Are you ready? Let’s get right into it.

person wearing lavatory gown with green stethoscope on neck using phone while standing — Photo by National Cancer Institute on Unsplash

Key Highlights

Why Fake Medical Images are important- Collecting and preparing high-quality data is very hard. This is doubly true for Medical Images. While there are a lot of people (thus data sources) in the world, very few people have the expertise to prepare medical data. Do you really trust a random Joe off the street to annotate a scan of your throat? Probably not. Therefore, high-quality medical annotations are very hard to come by.
The sharing problem- Medical Data itself comes with another issue- government regulations. Unlike a lot of other kinds of data, you can’t share medical data, even for research purposes. This makes the AI Data sharing Problem worse (learn more about that here).
How Synthetic Data solves both these issues- Using synthetic images (images generated using AI) can solve these issues. A good generator should be able to mimic human data, while also not having to worry about Privacy Laws.
Read my article on SinGAN-Seg to see the full results.
How Fake Medical Images are created- There are two components to creating these images. One, we need an AI generator that can generate human images. Next, to add more realism to these images, we use a technique called Neural Style Transfer (NST), created by Google. NST allows you to take the style of one image and the content of another image and combine them. The image below is an example-
In our case, we transfer the style of the real image to the content of the generated image. This is brilliant because it can help you avoid the weakness of using AI-generated images.
The Results- What were the results of this? Turns out we can match or even beat the performance of AI Models that use only real data, using synthetic data. Very exciting when you remember the implications.

To those of you that want to learn more about the architecture and why it works so well, check out the following video I made on it. I made it a while back, so my camera presence was very noobish.

To those of you that want ideas to take this to the next level, you can use randomness to add even more to this. Check out my LinkedIn post on integrating Randonmness into Machine Learning to learn more.

Sponsored Segment

AlphaSignal is a free weekly Summary of the top developments in Machine Learning. They use AI to rank and send you the top developments in the field. Check them out. Reading them is a great way to stay in touch with the field and support my writing.

https://alphasignal.ai/?referrer=Devansh

If you are someone interested in sponsoring this newsletter, let me know.

I created Technology Made Simple using new techniques discovered through mentoring multiple people in top tech firms. The newsletter is designed to help you succeed, saving you from hours wasted on going through substandard resources, the Leetcode grind, or multi-hour-long lectures. I have a 100% satisfaction policy, so you can try it out at no risk to you. You can read the FAQs and find out more here. Use the button below to get 20% off for up to a whole year. Using this discount will drop the prices-

800 INR (10 USD) → 533 INR (8 USD) per Month

8000 INR (100 USD) → 6400INR (80 USD) per year

Get 20% off for 1 year

Before proceeding, if you have enjoyed this post so far, please make sure you like it (the little heart button in the email/post).

In the comments below, share what topic you want to focus on. I’d be interested in learning and will cover them. To learn more about the newsletter, check our detailed About Page + FAQs

If you liked this post, make sure you fill out this survey. It’s anonymous and will take 2 minutes of your time. It will help me understand you better, allowing for better content.

https://forms.gle/XfTXSjnC8W2wR9qT9

I see you living the dream.

Go kill all and Stay Woke,

Devansh <3

To make sure you get the most out of Technique Tuesdays, make sure you’re checking in the rest of the days as well. Leverage all the techniques I have discovered through my successful tutoring to easily succeed in your interviews and save your time and energy by joining the premium subscribers down below. Get a discount (for a whole year) using the button below

Get 20% off for 1 year