Python Forum

Full Version: What type of Artificial Intelligence for Detecting Early Signs of Diseases via Image
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
What type of Artificial Intelligence for Detecting Early Signs of Diseases via Image Analysis?
I am seeking to build an Artificial Intelligence via Python that recognizes patterns in detecting early signs of disease, as the goal to prevent the disease. I want to use Supervised Learning. I was wondering if I should use Classification or Regression. The images I input are basically images from for example X-Ray where I have input in the information in the potential early stages of a patient's disease, about whether it is a disease or not. And since it's supervised learning, I have also output images of diseases and non-diseases which the machine trains on. In the end, I want to have the AI know from analyzing a picture whether it is a disease or not with great accuracy.

At the moment I am doing research. What concepts, subjects should I focus on?

Also, how much computing power is required for this? For example, if we have 50.000 pictures to analyze. I am trying this in Jupyter Notebook. Are there any more important factors that I need to know in order to make my project a reality? I am a beginner :) ANY help, tips would be APPRECIATED!
First, recommend classification, and convolutional neural networks.

This is already being done to a great extent with mammography - computer aided diagnosis. Skin lesions (melanoma) has been shown to be better diagnosed by algorithm than by dermatologists. Those are two areas I would look at to get familiar with the concepts.

I would use Pubmed to look up the publications regarding these areas, and other areas of interest to see what approaches are preferred, and what hurdles they have had to overcome.

https://pubmed.ncbi.nlm.nih.gov/

A list of related articles can be found here (pubmed search)

50,000 images may or may not be a problem, depending on your computer and the resolution of those images. I would give consideration to using colab from google, which is online, and where you can use GPUs and TPUs. On a project I did it would not run on IBM Watson (at least the free version) but ran fine on colab with GPU.

One thing you will want to decide upon is what your overall goal is - publication, advancement in your career, commercial product, etc. as your approach may differ. I'm happy to help.

Jeffrey A. Summers, MD FACP FSSCI
Professor and Chair, Department of Internal Medicine
Quillen College of Medicine
East Tennessee State University

(don't usually publish that here, but in this case it applies)
Thank you for your intelligent message.

Right now I am mapping what needs to be done and how.
My goal is to help humanity forward as much as I can.
I have a company called AIify, our mission is to integrate AI into every business.
But Healthcare especially is special for me.
So I think all of your mentioned goals can be applied here.

With the motto: Prevention is better than cure
I try to create artificial intelligence that can detect early signs of disease.
I am researching and asking as much as I can to have a clear path in front of me to take.

I have no background in coding, Artificial Intelligence, or Healthcare.
That is why right now I am categorizing what I need to learn in order to do this. Also what platforms to use, what coding language, what libraries, etc.
I learn this by Google search and asking questions on great forums like this.
But I do have a very strong passion for helping humanity forward, and what could be more important than saving lives!

That is why I am putting all my energy into improving the technology for preventing diseases since there will be no disease in the first place when prevented.

Thank you for giving examples of where Detecting Early Signs of Diseases with Image Analysis is applied, it gives me an idea of where to look for.
Also thank you for clearing out if I should use Classification or Regression and advising me to use Convolutional Neural Networks.

If you know anything that could help me out, please let me know. All help is appreciated :)
Out today - use of a convolutional neural network on EKGs to determine risk of a condition that predisposes to sudden death.

MedPage Today