Two of our team members are medical students and were on a medical mission trip to Honduras to perform cervical cancer screening. What we found though was that many patients were lost to follow-up due to the two-to-three week lag time between screening and diagnosis. We sought to eliminate this lag time and thus enable instant diagnosis of cervical cancer through Cerviscope.
What it does
Cerviscope is a product that enables point-of-care detection of cervical cancer based on hand-held light microscopy of pap smear slides and subsequent image analysis using convoluted neural networks. See our demo at www.StudentsAtYourCervix.com! (If URL forwarding doesn't work, https://www.wolframcloud.com/objects/ecb2d3d2-c468-4c4b-80fd-3da5f2e2aa90)
How we built it
Our solution is two-fold. The first is using machine learning, specifically convolutional neural networks, to analyze images from a publicly available database of annotated pap smear cervical cells to be able to predict cancerous cells. We couple our algorithm with the use of a handheld, mobile phone-compatible light microscope called the Foldscope, developed by the Prakash lab at Stanford, to enable cheap and rapid point of care detection. These images were not normalized, so we performed image processing and conformation tools to standardize these images as inputs to our neural network.
Challenges we ran into
1) Time. We would have loved to fully automate the interface between image acquisition using the hand-held microscope and machine-learning-based image analysis. 2) Data. Acquire a larger database to improve our accuracy closer to human level performance
Accomplishments that we're proud of
We trained a neural network to detect cervical cancer cells with classification accuracies exceeding 80%. Coupled with the utilization of the Foldscope, this has the potential for huge impact in the point-of-care cervical cancer diagnosis and treatment field.
What we learned
How to develop, train, and deploy neural networks in just a few hours. Network with students to learn about amazing work such as the Foldscope!
What's next for CerviScope - Cervical Cancer Detection
Sleep. Krispy Kreme. Then IPO.