X-rays - Radiologists

Deep Learning
1

Hello All

We are planning to build an affordable product to predict abnormalities in various body parts by screening X-rays. Can some of the radiologists in this community throw some light on the benefit of using such a product especially in developing countries where the number of radiologists are few and X-ray is one of the main sources of diagnosis?

Appreciate your help!
Thanks
Vineet

2

Hi Vineet,
I am not a radiologist, but worked for a meidical equipment companies for some years. I think doctors & radiologists can be benefited in two ways:

  • Released from workload pressure
  • Learning from starte-of-art detection methods.

Patients also can take some advantages:

  • More accurate detection. Because the medical quality in some developing countries is extreme unbalanced, especially for poor neighborhoods there are a small number of qualified practitioners.
  • Reducing waiting time. Since machines tend to be more effective in processing images.

However, there are two major problems prevent using of automatic x-ray products.
The first one is that, in many cases, doing X-ray is not enough to diagnose dieases directly, thus there is a little chance for deep learning to take effect ( disease filtering is ok).
The other thing is that cost control of such a product is extreme important. In developing countries, the salary of radiologist is not as high as in developed countries, hence if a product equiped with deep learning algorithms is more expensive than hiring a full-time radiologist, hospitals will probably choose to hire a new employee.

1 Like
3

Hello Kevin

Thank you very much for the inputs. I agree that X-rays is not ideal for diagnosis of diseases directly like cancer but we would like to start with chest X-rays to diagnose some of them like TB etc. Then, once this kind of product gains interest, we will simultaneously build CT diagnosis solutions.
We plan to roll out the product as a very affordable solution and there are not enough radiologists in India (1:100000) hence there is a problem of scaling up anyways.
Although, I agree that we need to make sure the product is a real value-add as per your suggestions. If you have any other thoughts, I would be happy to discuss.
More thoughts generates more solutions.

Thanks

4

hi Kevin

Can we achieve 98% accuracy for Chest X-rays? Do we have that data?

Thanks
Vineet

5

Vineet, you may find this website interesting: https://lukeoakdenrayner.wordpress.com

6

Thank you Mattias. It seems that there is a lot of things to consider for accuracy.
Although, what I am trying to build is a basic model which will augment radiologists in India where they are so really burdened with loads of scans. Even a basic model which will predict to a certain accuracy level is a good starting point I assume.

Thanks
Vineet