Hello everyone, I want to enhanced the medical image of skin cancer as well as feed the enhancement image to CNN for classification.
I wonder if generative adversarial network (GAN) is the best method? And how can I use GAN for the purpose of enhancement instead of generate a fake image.
One of the big advantages of Deep Learning for classification is that the model itself develop the features that leads to the best performance based on the training data.
It was found that for cellular image segmentation, color normalization was not very useful so I would be even more skeptical of "image enhancement" for skin cancer detection...
https://www.frontiersin.org/articles/10.3389/fbioe.2019.00300/full
Can you make a sample image available to do some enhancements and classification using algorithms and methods used in the remote sensing community? If so, I will take a look and generate some sample enhancements for you to view.
Pat
Thank you. Can you tell me a little about what is of interest in the image?
Pat S. Chavez, Jr. I want to reduce the image size while keeping the image resolution from degrading as much as possible
Enhance is a generic term, it says nothing concrete, so it is not really possible to answer your question. In comments you say "enhance the compression image": if compression degrades the image, the obvious enhancement is to avoid compressing.
Using Generative Adversarial Networks (GANs) for medical image enhancement can be beneficial in various applications, such as improving image quality, denoising, super-resolution, or synthesizing missing information. Here's a general approach to utilizing GANs for medical image enhancement:
Sorry it has been a while since I said I would ‘play around’ with your image to do a couple of enhancements (I have been tied up the last six weeks with some training). First, I want to say I have absolutely ZERO background in the medical field and am just guessing at what information might be of interest to you. I hope that I am correct in assuming that perhaps you are interested in information within the ‘darker area’ in the middle of the image.
The attached images represent:
· Original with a simple contrast stretch.
· Filter to enhance locate information.
· Ratio to enhance some of the colors within the dark area.
· Simple segmentation of one of the bands.
Let me know if you have any questions and or comments let me know.
Pat
Pat S. Chavez, Jr. Thank you very much for your answer, it's very kind of you and sorry to bother you with my question, and thank you very much for the slideshow