MOCSA: Mobile Oral Cancer Screening Application

The Project

The purpose of my web app is to help detect oral cancer in its early stages, by giving people a reliable and convenient way to determine whether an oral lesion in their mouth is benign or malignant, only through images taken by a mobile camera. This web app is directed towards people living in rural or underdeveloped communities who don’t have immediate access to a dentist. Instead of using histopathological images in my app, for which lab testing is required to obtain, I used only images taken by a mobile camera, which is much more convenient in this day and age. I created a machine learning model that would be able to distinguish between and categorize the images as either benign or malignant. To create a machine learning model with the highest accuracy possible, I tested the Convolutional Neural Network (CNN) models: MobileNetV2, Resnet50, and VGG16. I then altered two hyperparameters on each of these models: epochs and learning rate. Resnet50 had the best accuracy of 92.98% with the learning rate 0.00005 and epochs 90, while VGG16 had the best accuracy of 89.47% of with the learning rate 0.001 and epochs 20. Overall, MobileNetV2 had the highest accuracy of 98.237% with the learning rate 0.0005 and epochs 90, so it was the model I deployed into the app. In addition to the prediction dashboard, my app has an instructions page that tells users how to use the app and an information page that provides reliable information and the risk factors of oral cancer. There is also a disclaimer included that warns users about the possibility that the output they receive is incorrect, despite the high accuracy. After the user inputs their image, there is a next steps and resources page that pops up, providing them important information and links. My aim is to increase accessibility to dental health and to raise awareness about this topic. Web App: https://mahimaw27-mocsa-app-z3p5oi.streamlit.app/ App code: https://github.com/mahimaw27/MOCSA

View this project on YouTube

Team Comments

When I visited our village in India last summer, I saw that many people were chewing tobacco, which is a risk factor for oral cancer. They were not aware of this and didn’t have access to dental care either. I wanted to help people in these conditions by using my passion for coding to create MOCSA.

When making my project, my main issue was finding a dataset. I reached out to many researchers who had access to these types of datasets, but was continuously declined. In the end, after 3 months, I found a dataset on Mendeley data that fit all my criteria and had enough images to work with.

If I had more time, I would create a better version of the app with more accessibility features. I would also make the app available in multiple languages and add a map that shows the nearest dental facility to the user. I also want to put the app on iOS and Android so it can reach a wider audience.