The wonders, and the uncertainties, that come out of artificial intelligence (AI) spring from a common source: the ability of machine-learning algorithms to detect patterns. From astoundingly large stockpiles of data — beyond the ability of humans to wrap our minds around — AI turns up some sort of pattern.
That same facility enables devices to, say, translate between languages or navigate via GPS. It also creates promising opportunities to break ground in addressing cancer and other life-threatening diseases.
Exploring those possibilities is the domain of the City of Hope® Applied AI and Data Science team. They collaborate with clinicians on machine-learning strategies designed to offer better outcomes from surgery and improve patient care with predictive algorithms. The scope of their work includes many practical applications that free up providers so they can focus on patients, as well as the hunt for breakthroughs with the potential to alter the landscape of cancer treatment.
“We want to use automation to handle some parts of the job that are repetitive and mundane, so that doctors and nurses can do what only humans can do and perform at the top of their license,” said Nasim Eftekhari, M.S., executive director of Applied AI and Data Science at City of Hope. “Our projects that improve efficiency and decision-making help save lives. And then, maybe one in 100 projects is a moonshot. Some of those are very exciting.”
Eftekhari highlights three data science endeavors in particular that show how City of Hope is harnessing AI’s pattern-recognition superpower in search of innovations that make a difference for patients — aiming to detect cancer earlier, to assist oncologists and to enhance treatment planning.
A New View of Cancer’s Genetic Footprints
Genes meet generative AI in one effort underway at City of Hope. The better we understand cancer, the better we can find it and fight it, so the researchers focus their algorithm on the changes to how genes are expressed as a cell goes from noncancerous to malignant.
They aim to shed light on the stages in between health and cancer. The technology modeling the biochemical interactions along that path is best known from a more lighthearted use case: phone apps that start with a photo of a person and simulate what the progression of age might do to their looks. Results generated by the gene expression algorithm have been successfully tested against what’s currently known. It may just help map out the molecular progression toward disease.
“If we can learn more about what happens to gene expression before cancer, we may eventually be able to detect cancer earlier, or even before it happens,” Eftekhari said.
Getting to the Heart of Patient Medical Histories
As a National Cancer Institute-designated comprehensive cancer center, City of Hope treats many people who have a history with cancer and are seeking the best help they can find. Some have records spanning decades and including documents from different organizations, totaling hundreds and hundreds of pages.
City of Hope data scientists are working with doctors to condense such extensive histories, making it more straightforward to understand complex cases right from the start. Behind the scenes is an algorithm, trained on millions of oncology-specific data points, that summarizes a patient record in a clear-cut timeline. The summary is then passed to ChatGPT to clean up the report, doing things such as standardizing grammar.
“We have taken what the chatbot does best and guarded against its limitations,” Eftekhari said. “Our system takes more information than it’s humanly possible to go through in a short amount of time and turns it to something that can be reviewed in a page or two.”
Data-Driven Guidance for Treatment Planning
The electronic health record is a substantial boon for AI-based investigations. With medical histories digitized, they can be quantified and analyzed in new ways. One project exemplifies how anonymized patient data can be a promising resource for researchers looking to improve care.
The algorithm under development correlates details about one patient to the full store of data available on patients. It’s a comprehensive comparison that brings together everything from demographic, geographic and socioeconomic factors to genomics, tumor typing, clinical imaging and sensor data. Generating similarity scores, the system would return other cases that have the most in common with that individual’s. His or her physician could then use knowledge about what worked previously for closely matched patients to inform recommendations for therapy.
“Oncologists may call senior colleagues for advice, and we’re trying to replicate what they do,” Eftekhari said. “Great doctors connect the dots. They think about their experiences, about similar patients they’ve seen in the past. On a larger scale, we want to give all our doctors the benefit of City of Hope’s collective experience, right in their pockets.”