Artificial Intelligence and Cancer Diagnosis: Meet the 2012 Google Science Fair Winner
According to senior Brittany Wenger, programming computers to think for themselves in the name of science may change the face of cancer diagnosis. Using cloud technology to both share data and feed her program, Wenger's Google Science Fair project blends computer science and medicine to create a powerful resource for the medical community—and an inspiration for other young programmers.
Brittany's winning Google Science Fair project involved the creation of an artificial neural network (ANN) (and a cloud component for global access) to help with the evaluation of breast cancer biopsies. An example of artificial intelligence, an ANN is a program modeled after the biological neural networks of the brain. With training, these math-based networks can detect and synthesize incredibly complex patterns in data—and can continue to refine their accuracy, autonomously, as they analyze increasing amounts of data. "My network is a back propagation neural network, meaning it learns and updates its formula based on experiences and mistakes," explains Brittany, "so with more data and experiences, the network is able to fine tune itself better."
Top Science Student
A senior in Sarasota, Florida, Brittany appears at first glance to be a typical high school student. She plays two varsity sports and competitive soccer, takes a rigorous academic course load, and has accumulated more than 500 hours of community service. Look a little more closely, however, and you see that she is also an astute computer scientist with an unquenchable thirst for science. She is also, by necessity, a master of time management, and, when it comes to choosing between her research and day-to-day activities, she often winds up in the computer lab, especially during Fall months when her projects are at breakthrough points. "I was researching on Halloween," she admits, noting that, already at eighteen, she sometimes makes social sacrifices in the name of research. But she is pragmatic. "The Friday nights I give up are my choice. I love researching and do not mind some of the things I have had to give up. You can't do everything."
Brittany's research doesn't always leave a lot of room for sleep, either. At one point during her Google project, while in the process of running 7.6 million trials, Brittany had to set her alarm every four hours for two weeks in order to start new cycles. With or without sleep, science is a passion for Brittany. "Through science I can find answers—and more questions," says the Google Science Fair winner who began competing in science events in the sixth grade. Since then, she has accumulated a lengthy list of awards and successes at numerous competitions. As an eighth grader, she went to the Science & the Public (SSP) Middle School Program (equivalent to today's Broadcom MASTERS) and placed third. She has also competed and won awards at the Intel International Science and Engineering Fair (Intel ISEF) three years in a row. And then there was the 2012 Google Science Fair, where her persistence, passion, and sophisticated cross-disciplinary project wowed an impressive team of judges and took top honors.
First Steps in Computer Science
Initially interested in web-based information, especially online gaming and community sites, Brittany focused her first computer science projects on issues related to usability and user interface design. "My first science fair project explored whether people prefer a text, bulleted, or picture layout for a webpage," says Brittany. "My second project was on what age groups prefer what types of layouts (maps, menus, or plain text)." Once she got hooked on the idea of artificial intelligence, however, Brittany's interest kicked into high gear. "I bought a coding book and decided that that was where I was going to focus my energies." Armed with a book that walked her through basic computer programming examples, Brittany shifted from HTML scripting to more robust programming languages like C#, and the course of her computer projects and scientific research changed.
She began her first artificial intelligence project in the seventh grade. That project, her soccer neural network, was three years in the making, but in those three years, Brittany learned a tremendous amount about ANNs and gained hands-on coding experience that set the stage for her high school breast cancer neural network project. "My love of AI is deeply rooted in my experiences from my soccer program, and programming that first neural network was an integral part of my formation as a scientist," says Brittany. As a freshman, she presented her soccer ANN project at ISEF, where she won numerous awards, including a Special Award in Computer Science from Symantec and Science Buddies.
Women and Computer Science
Despite increased numbers of women pursuing careers in computer science, the field remains one steeped in stereotypes, a fact that, unfortunately, may dissuade female students from exploring even a fledgling interest in programming. Luckily, Brittany found support for her interest in computer science from the beginning.
Brittany admits that living in the age of the Internet, and with the benefits of open source code repositories, computer science is more inviting and accessible to a young coder. But having a real-world mentor and role model can be critical for any top science student. Brittany was fortunate to find excellent support in her high school's computer science department. "My computer science teacher, Mrs. Barrett, let me take AP Computer Science as a sophomore." This teacher pushed Brittany to explore Java, moving her beyond the comfort zone she had established with C#. But Brittany credits her success, overall, to all of her teachers. Brittany's project was interdisciplinary in nature, and she says she received invaluable support from her biology and math teachers.
Affecting 1 in every 8 women, breast cancer is a health problem that touches women of all ages. Many students have a family member or know someone who has had breast cancer. Brittany is no exception. As she began her project, a family member had just been diagnosed with breast cancer, a reality that gave her project added personal significance. Improving biopsy readings and helping ensure earlier and more accurate diagnosis has the power to change the grim statistics. It is on this premise and possibility that Brittany set to work building a system capable, she hopes, of making a difference in testing and diagnosis.
"When I found the University of Wisconsin's data published on the Machine Learning Repository and reviewed previous experiments, I was inspired to create a tool that could increase the success of studies using fine needle aspirates," she explains. "I wanted to create a tool that could be used to handle raw data and outliers while achieving sensitivity to malignancy. Such a tool will decrease the invasiveness of diagnostic procedures, decrease costs, and lead to earlier detection."
A Custom Solution
In preparation for her project, Brittany explored commercial solutions that are currently used for biopsy assessment. These off-the-shelf products allow users to customize artificial neural networks to analyze and act upon specific kinds of data, but they don't offer the specificity Brittany hoped to achieve with her own tool. As Brittany explains, using solutions like NeuroSolutions, NeuroIntelligence, and EasyNN, users can configure ANNs "to a degree," and without having to do any coding, similar to the way a user can use an office program like Excel to create and customize a graph.
As part of her benchmarking, Brittany configured each tool for "success with breast cancer diagnostics to their fullest potential." Then, with a hands-on understanding of the abilities and limitations of current tools, she built her own from scratch, creating a tool she says outperforms existing commercial solutions.
In every project, there are stumbling blocks, and computer programs require meticulous debugging and testing. Brittany's breast cancer ANN was no exception. "Global Neural Network Cloud Service for Breast Cancer failed twice before I coded a successful application," she says. "The first attempt, there were more errors than there was code, so I scrapped it without extensive troubleshooting." She made it farther with her second attempt. "It compiled, which was exciting at first, but then proved worse than flipping a coin at diagnosing breast cancer during testing."
Despite the setbacks, Brittany remained committed to her project and confident that she would create a successful tool. "In science, there are more flopped than successful experiments. Of course I wanted to create a viable diagnostic tool, but I also knew that with each flopped network, I was gaining valuable knowledge about what would—or more specifically what wouldn't—work with computer-aided diagnosis such as mine." Her persistence paid off as she, like her ANNs, learned along the way from each iteration of her project and research.
Brittany attributes the high accuracy of her ANN to three innovative factors: "I have an artificial input layer which converts each input into four nodes of its binary representation because binary spikes actually emulate the neural signals in the brain, since neurons are either firing or off. The service also has heavy malignant weighting, meaning it's going to err on the side of cancer because a false negative can be detrimental." A final differentiator between Brittany's solution and other products is the way her ANN determines and deals with what she calls "inconclusive logic." Rather than relying on a single determination from the ANN, Brittany's system takes into account ten different evaluations. "Instead of being derived from a sigmoid function, I actually simultaneously create ten different neural networks," she explains. "Since they all are randomly initialized a little differently, they all learn a little differently, just like all people learn a little differently. A sample is deemed inconclusive if these networks produce different diagnoses."
One of the key elements of Brittany's ANN involves how the system interprets and assimilates past "false negatives," biopsy readings that showed "negative" where, in fact, cancer was present. Brittany's system weights these assessments differently, setting her ANN up to learn from the existing data, and continue learning with the input of new data. "The weighting is not a 'guess,' but rather is adaptive learning," explains Brittany. "Initially, all the weights are randomly initialized as doubles (decimals between 0 and 1). The network learns via back-propagation, so these weights are updated and changed during training based on the network's experiences and mistakes."
A Cloud Approach
Brittany's ANN, alone, demonstrates powerful potential, potential she says can be used for other kinds of cancer diagnosis as well. But she didn't stop with the creation of the tool as a standalone application. Instead, she turned to cloud computing to make the ANN accessible to medical and research professionals around the world. "The cloud is an amazing elastic entity that allows me to create a tool accessible to the global medical community," says Brittany. "Right now, the service has low usage, but the cloud allows it to scale to support every hospital in the world if they want to access it tomorrow." The scalable and flexible nature of a cloud service also offers Brittany the ability to configure her program for use with a range of devices.
The Google Science Fair Experience
As a computer scientist, Brittany admits that when she first read about the Google Science Fair, she was intrigued by the new competition but also very excited about the possibility of visiting Google. She worked on her winning Google Science Fair project for two years, first entering an early iteration of her winning project in 2011. Though she was not a finalist that year, her experience with the virtual science fair was encouraging, and so she returned again with a more robust version of her ANN project in 2012.
"I really like the format of the Google Science Fair. All entries are submitted as online websites, which allows for participants around the world to submit entries," says Brittany. "I worked very hard on my site and did some HTML, but the Google tools and formatting ensure that even if you aren't a coder, you can have a good site." According to Brittany, the various sections required for a Google Science Fair project are clearly explained and adhere to the scientific method, which makes preparing a project submission a straightforward process.
As a finalist, Brittany joined other finalists for four days at Google. While she enjoyed all aspects of the visit, including the chance to ride in a self-driving car, Brittany was especially excited to meet and present her project to scientists like Ada Yonith (a chemistry Nobel laureate), Steve Myers (CERN/Higgs Boson), and Vint Cerf (one of the fathers of the Internet and vice president and chief Internet evangelist for Google).
As the winner of the 2012 Google Science Fair, Brittany has found herself in an enviable position—everyone wants to help. "I am so overwhelmed by the support I have received from my community and internationally. All the publicity has led to doctors contacting me, interested in collaborating so I can get more data." This collaboration will help Brittany continue to refine and educate her ANN, and as she feeds in more and more data, the ANN will continue to learn. "I believe that winning the Google Science Fair has given my research the opportunity to go to the next level, and possibly make a difference in actual patient's lives."
Even as she looks ahead to the next steps in her research and development, Brittany first gets to enjoy the enviable bounties of winning the Google Science Fair. "I get to go to the Galapagos with National Geographic and be inspired by the same islands as Darwin was. I get to have an experience with LEGO, Google, or Cern. I get to speak at breastcancer.org and at TED. As a scientist, these are experiences that are beyond my wildest dreams."
Life is busy for the high school senior. Since speaking with Science Buddies, Brittany gave her first TED talk at TEDxAtlanta and will be speaking at TEDxWomen later this month. She was part of the "The World We Dream" Google Zeitgeist panel, and attended the Equal Futures Partnership launch at the UN General assembly. In addition, she has been named a regional finalist in the Siemens Competition in Math, Science & Technology.
Without a doubt, Brittany's success is a shining example for young women who are drawn to computers at an early age. Finding female role models in computer science is incredibly important and empowering and can be critical in encouraging and nurturing early and continued interest. As the story of Brittany's evolution as a scientist and a programmer shows, even during elementary school years, girls can create computer science projects that fit their interests and let them develop core programming skills upon which they will continue to build as they progress to more complicated projects and programs. "I think my biggest advice for women interested in computer science would be to just go for it! I know there is a gender disparity in computer science, but it is not indicative of boys being better suited for the field than girls. I think it is important for young women to realize that coding is so much more than creating the hardcore video games that seem to have more appeal to boys than girls. It's our Internet, our social networking, and so many other technologies that do appeal to women and men equally."
You Might Also Enjoy these Previous Entries:
- Online Support Forum Helps Student Succeed Two Years in a Row
- Student Cleans Up Oil Spill Science with Support from Online Mentor
- Student Explores the Connection Between Human Behavior and Cybersecurity
- Sparking Interest in STEM Careers in Middle School
- Student Success and the Science of Stealth
- A Pet Science Project Success
- Science Buddies Teams up with PPG Volunteers to Improve Crime Scene Chemistry Project
- Inspiring Students about STEM Careers and Robotics