The human eye is not very accurate when it comes to detecting subtle changes in skin patterns and irregularities. While our eyes can recognize broad and obvious visual cues, they often struggle to discern the early signs of potential health issues, such as skin cancer. In the case of Yair's grandfather, this limitation led to a very late diagnosis of skin cancer, as he did not know that he had melanoma - he thought it was just moles that got more common as he aged. Luckily, his grandfather survived, but this experience fueled the inspiration for us to bridge the gap between human perception and advanced technology. With new technology such as this, we can revolutionize the medical field, one step at a time. This project is one of those steps, and it can be expanded/applied to other medical diagnoses as well, not just skin cancer.
🤔What it does🤔
SkinSight takes an input picture of an irregular spot of skin from the user, then analyzes it and detects whether or not that irregularity may be cancer. It outputs the top three most likely diagnoses and their respective analyzed percentage probabilities.
🔨How we built it🔨
- Built using Flask
- Python for backend, used OpenCV for image processing, used both Pytorch and Tensorflow for AI ML
💪Challenges we ran into💪
- Slight inaccuracies (just under 90% accuracy) due to not having a large enough database, however, the current database is already 100gb. The accuracy could be increased, but the database being used would have to expand a lot, and we do not have the means to get access to a server powerful enough to handle it.
- Not enough time to fully train AI
- Python bugs during development that stopped the program from running
- Library dependency issues
- Display bugs on the web
🏆Accomplishments that we're proud of🏆
We are proud that our app works fully, and by extension, to have successfully harnessed the power of machine learning in order to develop this tool that allows users to take charge of their skin health. Witnessing the app in action, accurately analyzing skin lesions and providing instant result fills us with a sense of fulfillment, knowing that we have completed what we sought out to do.
Although SkinSight is primarily for skin issue diagnoses, the way we built this project can be extended to other things, which could affect the medical industry in the future, with ongoing development.
Moreover, we take pride in the positive impact our creation will have on the future. We learned many things and even bettered our own understandings on different types of skin cancer. This achievement strengthens our belief that innovative technology can and should be a force for good, making healthcare more accessible and promoting proactive health management. With SkinSight, we feel that we have made a meaningful contribution, combining technology with empathy for the greater good.
🎓What we learned🎓
While creating this project, we learned many things:
- First and foremost, we learned a lot about skin cancer and its different types and how they are diagnosed.
- Second, we learned more about computer vision and machine learning, as well as their challenges and complexities despite having used these tools before.
- Lastly, as with any group project, we had different ideas and disagreed on some things. However, in order to work optimally, we had to have good teamwork and chemistry, so this experience served as a reminder that by talking to each other and working together well, the development process could have been faster, leaving us with no time crunch. The moral of this is that teamwork is essential, as it is a group project, not three individual projects.
🔜What's next for SkinSight?🔜
There is always room for improvement, and as stated before, the way we built this project can be applied to other surface irregularities in the medical field that cannot be discerned by the human eye. Also, as stated earlier in the challenges section, SkinSight is not 100% accurate. Naturally, one of the next steps would be to increase the size of the database and give the AI more time to learn for maximum accuracy. In the future, we would like this to become a webapp that can be downloaded on phones to make it very accessible. More ideas for development include:
- Teaming with a local hospital or dermatologist for easy professional contact in the case that somebody has an irregularity with a high chance of skin cancer.
- Feedback feature from users, which could also be applied to help with the machine learning
- Applying the AI for other skin conditions
- Using the results for educational content exploration
By taking these next steps, we can ensure skin cancer detection continues to evolve, reaching more individuals in need, and making a lasting and positive impact on skin cancer awareness, prevention, and early detection.
Overall, the development of SkinSight was a fantastic learning experience for us, with a very valuable outcome! Thank you for reading through this project's description. Try it out yourself with the files linked in github below!