FourSight: Analysis of Cancerous Genetic Profiles With Artificial Neural Networks


Project Name

FourSight: Analysis of Cancerous Genetic Profiles With Artificial Neural Networks

Team Members

Koral Kulacoglu, Hanze Wu, Bryan Deng, Ethan Zhao, Alex Gan

School Name

St. Robert Catholic High School

Project Description

Early cancer detection is one of the most researched topics in the medical field. Yet, despite notable progress, 20% of cancer diagnoses are incorrect, with 28% of inaccurate or delayed diagnoses leading to death. The most common diagnostics rely on physical examinations, MRI or CT scans, all conducted by doctors. However, in recent years, biomedical technologies have allowed us to analyze the gene expression of an individual’s cells. One gene sequencing technique called microarrays can analyze an individual’s gene profile to spot various genetic abnormalities that indicate certain disease conditions. With this issue in mind, we built a microarray-based diagnosis AI for cancer identification. Using a massive dataset of 17,375 human miRNA microarray samples that we compiled together from the GEO database, we trained our program to 95% accuracy in recognizing cancerous genetic profiles. Our AI was able to remain accurate when tested with 12 different types of cancer. To make our innovation useful to the general public, we wrote a simple microarray diagnosis program, which recreates visual representations of the microarray inputs and reveals the AI’s certainty in each of its diagnoses. Upon competing in the Canadian Wide Science Fair, we received the national Platinum Prize and qualified for the Canadian National Team for the EUCYS competition in Leiden, Netherlands. The FourSight Team currently has a scientific paper being peer-reviewed in the CSFJournal, as well as being in communication with several professors and medical professionals regarding possible improvement and the implementation of our program in the medical setting. We hope to test out our program with clinical samples from undiagnosed patients in the future.

What Makes Your Project Remarkable?

We attempted to revolutionize cancer diagnosis by combining two of science’s newest innovations: artificial intelligence and genetic sequencing. Physical screening examinations are ineffective at detecting internal organ tumor masses, while biopsy surgeries risk accidental metastasis of tumor cells. The accuracy of our program is around 15% more accurate than conventional diagnosis methods (95% vs 80%) and the extraction of genetic sequences from the body is completely non-invasive, requiring only body fluid samples. With the success of our project, we could readily and accurately detect cancer at its early stages –diagnosing cancer before its development into later, fatal stages. Meanwhile, the complex nature of genetic tests, with over 2565 genes and intensity values, is incomprehensible for the human brain. Our program is able to make a diagnosis at a glance of a large set of numbers and not overlook nuanced details in the tes, unlike human professionals. Furthermore, using statistical methods, our AI was able to identify several biomarker genes responsible for different types of cancer. Using our AI, genetic cancer research can also be advanced at a rapid pace, as the AI is provided with more samples from which to learn. We hope to bring the non-invasive screening methods and the gene expression discovery potential of our free-to-use AI project into the public through a website currently in the works. In the future, we hope that we can bring more foresight into the field of genetic cancer screening.

Project Website

Project Video

Social Media Link(s)