Revolutionizing Cancer Diagnosis with SEQUOIA

February 11, 2025 – by Rebecca Handler

One of the hardest – and most expensive – challenges of diagnosing and treating cancer is time. Weeks can pass as patients wait for costly genetic tests to guide their treatment, and this is often time they can’t afford. 

But what if a simple biopsy slide could instantly reveal the genetic blueprint of a tumor, saving both precious time and thousands of dollars? Enter SEQUOIA, a groundbreaking AI tool developed by Olivier Gaevart, PhD, and his team at Stanford.

SEQUOIA – short for Slide-based Expression Quantification Using Linearized Attention – is a tool that can predict the activity of over 15,000 genes by analyzing standard microscope images of tumor biopsies. In addition to saving precious time and reducing costs, this technology enhances medical insight –  allowing clinicians to detect patterns and developments indiscernable to the human eye – and could significantly improve health outcomes for patients. 

Why Gene Expression Matters in Cancer Diagnosis 

Gene expression in a tumor reveals the biological processes fueling its growth – like inflammation or hypoxia – and provides clues about how aggressive the cancer might be or which treatments will work best. “When we use RNA sequencing for cancer patients, we get a snapshot of all the processes and pathways that are active in the tumor,” explained Gevaert. “This gives us valuable insights for diagnostics, prognostics, and even drug discovery.”

Traditionally, obtaining this information requires sequencing the tumor’s RNA, a process that involves extracting genetic material from tissue samples and analyzing it in specialized laboratories. This is expensive and logistically challenging, particularly in lower-resourced settings where access to genomic testing infrastructure is limited.

SEQUOIA eliminates the need for RNA sequencing, offering a faster and more affordable alternative by analyzing stained biopsy slides –images pathologists already use for cancer staging. “The image acts as a surrogate for what’s going on molecularly,” said Gevaert.

Gene expression can change the appearance of tumor cells, such as their texture, structure, or spatial organization. SEQUOIA’s AI model detects these subtle changes in high-resolution images, finding signals that are key to predicting gene activity and understanding the tumor’s biology but that are undetectable by the human eye alone.

Cutting Costs, Saving Lives

Using SEQUOIA, clinicians could bypass the need for molecular tests altogether. For breast cancer, the tool already shows promise. SEQUOIA can mimic gene-based tests like MammaPrint®, which helps determine whether a patient is at high risk of recurrence and whether they would benefit from chemotherapy. “We demonstrated that the AI predictions approximate the original molecular assays with similar accuracy,” Gevaert explained.

Beyond cost savings, SEQUOIA could significantly reduce waiting times for results, a critical advantage for patients with aggressive cancers. For example, while RNA sequencing can take weeks, SEQUOIA’s analysis is instantaneous once the slide image is obtained.

Building the Future of Cancer Care

Building SEQUOIA required integrating large datasets, including images and molecular information from more than 10,000 patients in public initiatives like The Cancer Genome Atlas.

Current AI predictor models analyze images which have been split into thousands of ‘tiles’ or sections of the tumor and have lacked the ability to identify relationships between the tiles in their analysis. SEQUOIA uses a type of AI architecture called a transformer, coupled with advanced modeling techniques, which can interpret all the tiles in relationship with another, thus allowing it to account for genetic variations within different parts of the tumor. This capability is critical, as tumors often exhibit significant heterogeneity—different regions of the tumor may have distinct genetic activity that influences how the cancer grows and responds to treatment.

 Transformers, a hallmark of cutting-edge AI, excel at capturing complex patterns and relationships within data. In SEQUOIA’s case, this means it can detect subtle genetic differences across different regions of the tumor, providing a more granular and accurate representation of gene activity. “When we started using these digital pathology foundation models, the performance of SEQUOIA improved dramatically,” Gevaert noted, highlighting how these innovations propelled the tool’s success.

Expanding the Horizon: SEQUOIA’s Potential Beyond Cancer

SEQUOIA’s applications extend beyond cancer. “The model isn’t tied to any specific cancer type or even cancer itself,” said Gevaert. It could one day be used for other diseases, such as infectious diseases, where digital pathology could play a role in understanding molecular pathways and guiding treatment.

While SEQUOIA is still undergoing research and validation, its potential is enormous. Gevaert envisions a future where AI-powered tools like SEQUOIA become a standard part of pathology workflows. “In a few years, we could see this AI tool deployed in clinical practice,” he said.

Beyond diagnostics, SEQUOIA could also play a role in drug discovery, using its insights into tumor biology to identify potential targets for new therapies. “The goal is to use this technology to make precision medicine more accessible, affordable, and impactful,” Gevaert concluded.