It is really hard for women to check their health conditions nowadays. Economical conditions or emotional situation they are in even they might simply forget all because of these breast cancer takes many lives. Most of the countries are not well developed or under-developed to provide medical devices to local clinics,hospitals which makes mammograms unaffordable,unreachable for women . Even some women is so shy to take this kind of examinations.
After making examinations doctors still are not hundred percent sure to diagnose patient with breast cancer.It is meticulous process and they should take into account all tests. So they need AI based assistant to help them.
To address all the problems indicated above we created simple solution solves all.
For doctors AI based mammogram analyzer & cell histopathology analyzer with high accuracy
For shy or emotionally sensitive women self-examiner
Email notifier for keeping women up to date about breast cancer
What it does (Purpose & How this application works)
Breast cancer is most commonly seen cancer in USA. In early stage symptoms might be neglected or are not noticed by women . So early detection is key part for preventing.
Early detection is crucial phase if cancerous tissue,lump or cyst detected at that stage it will lead to early treatment eventually can save a lot lives.
Mammo is an fully functional web-based app tackles the issue.It has 3 early detection test and techniques :
- Ai based Xray scanner
- Microscopic cell imaging
- Algorithmic decision maker
Ai based Xray scanner is tool uses custom deep learning network to indicate probabilistic cancer score of given xray image.Due to lack of training (about 2 hours) and training data (about 100 positive image) accuracy is not high.But still powerfull tool works well. It is complicated task to identify cancer on given xray image even for experienced doctors.
Microscopic cell imaging is tool uses deep learning techniques to serve better prediction.
Algorithmic decision maker is algorithmic tool by asking few questions calculates cancer score. This method is not most realible way identification of cancer but helpful for people from under-developed ,undeveloped or poor countries. Also can be used for monthly self assessment.
By email notifications app notifies user for monthly self-checkup , weekly tips about breast cancer and healthy lifestyle .
Eventually every doctor is responsible for what he prescribed,diagnosed and wrong diagnose might effect women mentally.This also effects doctors they want to be hundred percent sure everthing is just right and accurate.AI based this project can be used by doctors as well !
How I built it (How this application was developed)
Using pytorch,flask,css,html,bootstrap,pandas,numpy,opencv and etc Xray model is custom made that is why research and development of this project took a lot of time.
Challenges I ran into (Difficulties & Challenges you faced)
Training network with very limited amount of positive data (highly biased).Training took a while to complete. Some other data related and model output issues. There was only one dataset publicly available and it was very limited .I did my best in lack of data and lack of computational resource conditions.
Accomplishments that I'm proud of
Trained network with limited resource and data . Read several research data and created my network which performs well. Completed fully
functional app from scratch before the deadline.
What I learned
Working with severely biased data that i have never done before.
What's next for Mamma (Go-to-Market Evaluation (How will this product be made scalable and accessible to the public?))
There is two different possibility : as public and private app , public app as social responsibility project so every woman can use it examine themselves easily but as private app doctors or private clinics can use its AI based parts for better and accurate detection or identification of cancer. At this stage project is ready for deployment and production but still needs some retouch and accuracy increase for better prediction. After researching market my our conclusion is that there are just few start-up works this kind of AI based breast cancer detector and most of them are not publicly available. As a extension to current app mobile version can be made easily.