MelaScan

Portion of code that reads the inputted image to decide whether it is cancerous
States whether the skin lesion is Melanoma or not. This specifically states "predicted class is Melanoma"
Non-Melanoma picture is inputted.
Precision & Recall
The accuracy function.
The loss function.

Inspiration

The inspiration behind this specific project was our determination to make the detection of skin cancer more accessible to the common population. Healthcare tends to be expensive and daunting, for this reason, many are untrusting of the healthcare system. By creating this project we hope to provide easy access to early detection of skin cancer.

What it does

The user inputs an image of a skin lesion and the program identifies particular patterns in the image and outputs whether the predicted class is Melanoma or not.

How we built it

This code was written on an online IDE (jupyter labs) utilizing Python. We imported various libraries and developed a deep convolutional neural network (DCNN) so that it could identify patterns in the images of the skin lesions. Based on the patterns we wrote code that could output whether the skin lesion was melanoma or not.

Challenges we ran into

We were struggling to increase the accuracy of the program as the data was overfitting. Thus based on the DCNN the accuracy was around 76%.

Accomplishments that we're proud of

This was our first hackathon so being able to create a functioning program that focuses on improving diagnostic safety during the detection of skin cancer, was something we were really proud of. We wanted this to be a learning experience, so being able to make a somewhat accurate Melanoma detector was accomplishing.

What we learned

We learned how to develop a DCNN and to train such neural networks so that they can help with image recognition. Furthermore, we learned how to make a very user-friendly interface, to fulfill the user's needs in an efficient way.

What's next for MelaScan

We definitely want to improve MelaScan and make it into fully-fledged software that has a very high accuracy with a very user-friendly interface. We want this to be accessible to anyone so that preventative measures can be taken especially before it intensifies to metastatic melanoma (which spreads to lymph nodes, the brain, bones, liver, or lungs).