I just completed a draft copy of my thesis, but I still do not feel that it is up to par. I would need some input from those who have done research on medical imaging.
What are the common pitfalls to avoid?
How many pages and a bibliography?
How about clinical validation? Should I stick with majority voting or rely only on insights from a seasoned pulmonologist?
I look forward to your valuable insights.
@Victor Ikechukwu Agughasi,
I assume this is related to your paper, "ResNet-50 vs VGG-19 vs Training from Scratch: A comparative analysis of the segmentation and classification of Pneumonia from chest x-ray images"
So you are writing about "Issues in Validated Machine Learning for Automated Diagnosis Using Affordable Chest X-ray Images"?
Talk about the social, economic and human impacts of your research. You do not have to validate it now. Write your goals and issues in your thesis and use that to get your next job or assignment. AIs are NOT to be trusted without a certification and validation process. No matter how smart your machine algorithms, not matter how popular the tool might be, if there is no process for verification in place, it should NOT be used on humans.
You might want to recommend "global" (meaning all countries) challenge projects so ALL groups working on machine vision algorithms for diagnosis using chest x-rays can test their algorithms in open challenges. This is very likely happening already, but probably ad hoc and probably "gamed" to benefit a few. Seasoned practitioners can have a voice in developing "global best practices", if you help develop "open verifiable network projects" that cannot be gamed by any one group or country.
I suggest you throw in the term "sustainable" because if you try to force algorithm methods into traditional human only settings, and do not carefully plan for training and remote verification networks, you will hurt real people. You do not want to be responsible for people dying because your process is so slow or complex it takes too long. I say that because of where you are going to school. Because of your country. Workers can use software to help them in their jobs, humans can also supervise AIs to do their work. And humans and AIs working together can create new global industries so workers can work anywhere and live in a low cost country. If there are verification processes in place to assure quality and trace responsibility.
When you write your Thesis send me a copy. Best wishes.
I filed this in my notes under "Setting up Global Validation Networks for Machine assisted Medical Diagnoses and 'AI' Algorithms"
Richard Collins, The Internet Foundation
Richard Collins , thank you for the prompt response. To an extent, my research paper on "ResNet-50 vs VGG-19 vs Training from Scratch: A comparative analysis of the segmentation and classification of Pneumonia from chest x-ray images" is a subset of the thesis on AI model for COPD Diagnosis from CXR images for which some sections are dedicated to addressing issues in Validated Machine Learning for Automated Diagnosis Using Affordable Chest X-ray Images. As you pointed out "AIs are NOT to be trusted without a certification and validation process", and such involves extensive training and model verification which might take a couple of years to validate.
I will incorporate your valuable insights and will get back to you at the earliest.
Best regards,
Victor I. A.
Victor Ikechukwu Agughasi I just had the 25th Anniversary of The Internet Foundation on 23 Jul 2023. Again and again I see groups working on common topics, but NOT working together. Are there really more than 100 Million chest x-rays in India every year? But that is many separate efforts, and not one common project. Now in the world India is probably not a leader in gathering chest x-ray data into common formats, establishing best algorithms. Rather than cooperating with China and all other countries, groups will want to be "king of the hill" (a children's game, who is strongest). Global research is driven by jealousy and politics, self-interest and massive waste as groups duplicate what is already known. It is the data and models that are most important. If everyone works on their own piece of the puzzle, and does not share, everyone loses. Sounds like you have some good material in your thesis. Again, the thesis is just a step in what you might find is a very long lifetime of exploration and helping others. Try to write from your heart about real issues and what you hope to accomplish. If you think something needs to be done at national or global levels, say so and get people and groups moving in common direction. Be careful not to enable more monopolies and hoarding for self-interests. All I can suggest is so at least try to make sustainable open verifiable truly accessible systems. Then if the systems start to go astray, the tools for accepting feedback and implementing corrections will be there and work as they are supposed to. Systems should not be allowed to lock themselves tightly and ignore all feedback, which is more the norm now than the exception.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India