=======================

MEDICAL IMAGENET

Helping machines generalize medical imaging

Inspiration –

Theoretically speaking, ImageNet consists of Natural images, hence using Pre-trained models on ImageNet are generally helpful.There is a huge difference in natural images and medical images, And fine-tune of model is generally required, which is known as transfer learning. Occasionally, transfer learning may be no better than training from scratch, as the networks learn very different high-level features in the two tasks.

But what if we had pre-trained model on medical images, where model understands a bit of all types of images – maybe that could be a better Pre-trained model to use for this specific task.

Questions of interest –

• Explore the idea of generalized database like ImageNet in medical imaging. Is this a feasible thing to do?
• Is Fine-tuning from ImageNet to a specific tasks as and when required a better approach ?
• Is this even a solution we need? How good is transfer learning ?
• Does Transfer learning has any limitations, on types of data it can switch to?

First layout of idea-

Here is the initial idea i thought about –

“Medical IMAGENET”

Next Few Steps

• Has someone already done it before?
• If not, why? What are the inherent problems in this task?
• Is such a Dataset possible for medical imaging or is it too complex?

Articles to get you started –

• Understanding Transfer Learning for Medical Imaging
• A scoping review of transfer learning research on medical image analysis using ImageNet – Mohammad Amin Morid, PhD , Alireza Borjali, PhD, Guilherme Del Fiol, MD, PhD https://arxiv.org/abs/2004.13175
• https://www.researchgate.net/post/Possible_How_Transfer_learning_for_medical_images
• An overview of deep learning in medical imaging focusing on MRI – https://www.sciencedirect.com/science/article/pii/S0939388918301181#sec0095

How can you help?

• Literature review
• Talking to expertise and expanding this idea.
• Collect a list of potential Medical databases that can be used.

Are you knowledable about this topic, or interested in this idea ? To get involved – ping me!