Brittany Wenger has developed a sophisticated artificial neural network to detect breast cancer.
Brittany Wenger says that breast cancer statistics become much more than numbers when you know someone affected by the disease. She then tells the story of her cousin diagnosed with breast cancer and the impact that it has on a woman and her family.
In her TEDx Women talk Wenger discussed the processes she used to create her cancer detection algorithm, Cloud4Cancer. She focused on fine needle aspiration, a technique that is generally less invasive than biopsy but much less conclusive.
Her artificial neural network was modeled after the brain’s neurons and interconnections. The algorithm can detect patterns that aren’t visible to the human eye, and the network is constantly learning and adapting to new information. As cancer mutates and presents new characteristics the algorithm can also adapt to form new diagnoses.
http://tedxwomen.org/speakers/brittany-wenger/
Medical personnel give the algorithm nine different attributes from the fine needle aspiration test. Each of the attributes is then translated into four binary nodes. The hidden layer has six additional nodes that creates a full 216 different possibilities. The last process is the output layer of the program that gives the diagnosis.
The diagnosis application now exists in the cloud as CancerDiagnosisService and the CancerNeuralNetwork. Doctors can input their test results through the Cloud4Cancer app and then receive a diagnosis. Brittany says that by building the test as a web service it can be easily adapted to new technologies.
Wenger discloses that her algorithm acts like a brain – binary numbers are ones and zeroes and the brain’s neurons are either firing or not firing, on or off. The application works with raw unformatted data with no points removed by doctors.
Ten different neural networks process the data and all ten need to agree on a diagnosis or the result is labeled inconclusive. Current estimates say that the algorithm is 99.1% sensitive to malignancy and about 5% higher than commercial products. Wenger hopes that this means her system will soon be hospital ready.
The Cloud4Cancer app is demonstrated during the talk and a result is given immediately. 7.6million tests have been run by the algorithm and more data makes the test smarter and more accurate.
Brittany Wenger is a great presenter and she keeps the idea of scientific method threaded through her entire talk. She tells us that her failed versions of the cancer test taught her just as much as the successes. The talk is ended with one of my favorite questions to ask – why does this matter? Wenger’s reply is another question – who do you know that would benefit from better breast cancer detection?