AI meets oncology: New model personalizes bladder cancer treatment

(Press-News.org) Leveraging the power of AI and machine learning technologies, researchers at Weill Cornell Medicine developed a more effective model for predicting how patients with muscle-invasive bladder cancer will respond to chemotherapy. The model harnesses whole-slide tumor imaging data and gene expression analyses in a way that outperforms previous models using a single data type.

The study, published March 22 in npj Digital Medicine, identifies key genes and tumor characteristics that may determine treatment success. The ability to accurately anticipate how an individual will react to the standard-of-care therapy for this malignant cancer may help doctors personalize treatment and could potentially save those who respond well from undergoing bladder removal.

“This work represents the spirit of precision medicine,” said Dr. Fei Wang, professor of population health sciences at Weill Cornell Medicine and founding director of the Institute of Artificial Intelligence for Digital Health, who co-leads the study.  

“We want to identify the right treatment for the right patient at the right time,” added co-lead Dr. Bishoy Morris Faltas, the Gellert Family–John P. Leonard MD Research Scholar in Hematology and Medical Oncology and an associate professor of medicine and of cell and developmental biology at Weill Cornell Medicine, and an oncologist at NewYork-Presbyterian/Weill Cornell Medical Center.

Dr. Zilong Bai, research associate in population health sciences, and Dr. Mohamed Osman, postdoctoral associate in medicine, at Weill Cornell Medicine, collaboratively spearheaded this work.

Better Model, Better Predictions

To build a better predictive model, the two lead researchers teamed up. While Dr. Wang’s lab focuses on data mining and cutting-edge machine learning analyses, Dr. Faltas is a physician-scientist with expertise in bladder cancer biology.

They turned to data from the SWOG Cancer Research Network that designs and conducts multi-center clinical trials for adult cancers. Specifically, the researchers integrated data from images of prepared tumor samples with gene expression profiles, which provide a snapshot of the genes that are "turned on" or "off.” 

“Since expression patterns alone were not sufficient to predict patients’ responses in previous studies, we decided to pull in more information for our model,” said Dr. Faltas, who is also the chief research officer at the Englander Institute for Precision Medicine and a member of the Sandra and Edward Meyer Cancer Center at Weill Cornell Medicine.

To analyze the images, the researchers used specialized AI methods called graph neural networks, which capture how cancer cells, immune cells and fibroblasts are organized and interact within the tumor. They also incorporated automated image analysis to identify these different cell types at the tumor site.

Combining the image-based inputs with the gene expression data to train and test their AI-driven, deep-learning model, resulted in better clinical response predictions than models that used gene expression or imaging alone.

“On a scale of 0 to 1, where 1 is perfect and 0 means nothing is correct, our multimodal model gets close to 0.8, whereas unimodal models relying on only one source of data can achieve approximately 0.6,” said Dr. Wang. “That’s already exciting, but we plan to hone the model for further improvements.”

The Search for Biomarkers

As the researchers look for biomarkers such as genes that are predictive of clinical outcomes, they are finding clues that make sense. “I could see some of the genes I know are biologically relevant, not just random genes,” Dr. Faltas said. “That was reassuring and a sign that we were onto something important.”

The researchers plan to feed more types of data into the model such as mutational analyses of tumor DNA that can be picked up in blood or urine, or spatial analyses that would allow more precise identification of exactly what types of cells are present in the bladder. “That’s one of the key findings of our study—that the data synergize to improve prediction,” Dr. Faltas said.

The model also suggested some new hypotheses that Dr. Faltas and Dr. Wang are planning to test further. For example, the ratio of tumor cells to normal tissue cells, such as fibroblasts, impacts the response to chemotherapy predictions. “Perhaps an abundance of fibroblasts can shield tumor cells from chemotherapeutic drugs or support cancer cell growth. I would like to delve further into that biology,” he added.

In the meantime, Drs. Wang and Faltas will work on validating their findings in other clinical trial cohorts—and are open to extending their collaboration to determine whether their model can predict therapeutic response in a broader population of patients.

“The dream is that patients would walk into my office, and I could integrate all of their data into the AI framework and give them a score that predicts how they would respond to a particular therapy,” Dr. Faltas said. “It’s going to happen. But physicians like me will have to learn how to interpret these AI predictions and know that I can trust them—and to be able to explain them to my patients in a way they can also trust.”

END