The images needeed for diagnosis are dermascopy images which need to be taken by a dermatologist. Thus this isn't entirely remote, however by providing a more effective and accurate analysis this could cut down on the number of trips required to the dermatologist and can also be much more convenient and cheaper.
MelanomAI is still a much better alternative to current methods of diagnosis.
There are ways for people to do self dermascopy, but those ideas are still in development. (This could be something we do in another hackathon or after this hackathon to suplement and support this idea)
After the hackathon we worked on it a bit more and were able to host it. However due to the size of the ML models we were unable to deploy those. You can still enter an image, your email, and get a message demonstrating how the website would function.
Check it out: https://melanomai.herokuapp.com/
Melanoma is a deadly skin cancer which affects all ages. It starts off as a cancerous growth but can spread to other parts of the body as well.
The worst part is that Melanoma has a 25 - 30 percent misdiagnosis rate meaning 1 in 4 people have been misdiagnosed with the cancer.
Considering how dangerous and scary cancer can be a 25% misdiagnosis rate is too high and 1 in 4 people should not have to suffer due to an accident that can be avoided.
More so with the current Covid Situation, remote diagnosis is as important as ever as a extra few trips to the hospital could mean the difference between life and death.
This leads us to the question, how can we accurately and remotely diagnose melanoma?
MelanomAI is the answer to this problem.
What it does
MelanomAI analyzes the image of suspected melanoma to detect whether or not it is melanoma. Once the analysis is complete the user gets and email confirming their results where they can then seek out the proper help based on the diagnosis.
With MelanomAI, a diagnosis is just 5 steps away, and the need to go to a dermatologist or run multiple tests is severely lowered. The AI is on par, and sometimes even more accurate than current healthcare professionals which means that users can trust the AI's results.
The best part is that it doesn't require a checkup, testing, or other measures. Its as simple as uploading an image.
How We built it
We used pytorch to build the AI model and train it.
We used bootstrap, django, css, and html to develop the website.
Challenges I ran into
One of our members lost half of their files during a crucial time and had to recode all of them. It was a traumatic experience and was a good learning opportunity on how to store files and use github. More so, integrating the AI into the website prove to be more challenging than expected, but after some troubleshooting we were able to make it work!
Accomplishments that I'm proud of
We are proud of how we were able to create a working website and integrate the AI into it. In previous hackathons we were unable to integrate both the AI and Website, but this time we managed to do so which was a large milestone for us.
What We learned
By developing the website we learned how to use django and integrate AI into a website. We also learned more about AI and how to use pytorch to make AI models.
What's next for MelanomAI
We want to refine the algorithim to work more efficently and predict more accurately. As of now it seems to have some bugs, but given some time we are confident we can work them out and increase our accuracy to close to 100%.
After refinement, our next step would be to host our website on a powerful server that can handle deep learning models. We hope to accomplish this task in the oncoming weeks and are excited to make this a reality!