During my 10 days at the TeensInAI Accelerator I’ve learnt a myriad of skills from presenting, communication to programming and design thinking. I’d like to thank Elena Sinel for organising this event and giving all of us the opportunity to come up with such creative and unique solutions to world problems, but also believing in us every step of the way.
My team decided to tackle breast cancer, however, we originally started tackling homelessness, but, because there wasn’t enough data, we had to pivot 4 days before the pitch. Despite this drawback, we managed to create a product using a neural network and a GUI. My role in the team was the developer, and even though I knew how to code in Python, I had never used neural networks or any form of machine learning in Python — so I had to learn very quickly. However, I wouldn’t have been able to learn so much without the help of the amazing mentors in the programme.
Our brand name is EarlyCatch and our aim is to use image classification and deep learning to help diagnose breast cancer sooner using FNA samples.
When we initially tackled this issue, we found data first because we didn’t want to be left in the same position as our last endeavour; we could not create our product as we didn’t have enough data. We found approximately images that tested both positive and negative for breast cancer (approximately 2,500 for each). Despite the small amount of data that we had, we decided to use it as we did not have another option. From a technical perspective, we started using a logistic regression algorithm to fit the data best, giving us a 64% accuracy. However, we needed a higher value for our product to be viable on pitch day. Therefore, we decided to use a Support Vector Machine (SVM), which gave an even lower accuracy of 48%. With the help of a mentor, we decided to use a linear neural network and it gave a 78% accuracy — the best result that we could achieve with the resources that we had. Originally, we had planned on making a convolutional neural network (CNN) and training it ourselves. However, this required large amounts of processing power, which we didn’t have.
The past 10 days have been an amazing journey and I have learnt a lot, I’m very grateful for being a part of this programme.
CTO of EarlyCatch