Skin cancer is a major public health problem, with over 5 million newly diagnosed cases in the United States each year. Melanoma is the deadliest form of skin cancer, responsible for over 9,000 deaths each year. Catching it early greatly increases chances of survival and enables removing it. It’s primarily diagnosed visually. A mole can be Benign(Just a mole) or Malignant(Cancerous cells).
What it does
With this project, we aim to develop an application that will help a user quickly and easily identify whether the moles on his skin can cause melanoma or not. Using the mobile (multi-platform) users can scan their moles and get a hint about consulting a dermatologist.
How we built it
Some of the technologies we used includes:
- Python and Flask Framework
- Deep learning and Tensorflow
- Google Cloud and Amazon AWS
Challenges I ran into
Dealing with huge image-based dataset is not easy when it comes to deep learning. Google Cloud platform helped a lot in training the model quickly using a GPU-enabled instance.
Accomplishments that we are proud of
We are proud that we completed the project on time, reaching a good accuracy in classification of skin lesions.
What I learned
We learned how to quickly prototype a mobile multi-platform app which accesses the camera and the photo library and communicates to a backend using rest APIs. We learned how to quickly spin a RESTful service able to run on the cloud enabled to use serverless technology. Most importantly, we learned how to train a deep neural network to classify images using Tensorflow implementation. We handled a massive image dataset and used GPU-enabled Google Cloud instances to run our training phase.
What's next for DermaScan
DermScan can be extended to any skin rash, i.e. psoriasis or eczema. Moreover, improvements can be done in terms of accuracy as it is still an open problem in research. Moreover, DermaScan can be directly linked to a healthcare center which can automatically receive images of suspicious moles and provide dermatology consultations to have a more effective impact on the patient.