A deep learning-enabled precision medicine approach could help predict sex-specific patterns of glioblastoma growth, which might improve risk stratification and treatment planning.
Researchers from the University of Wisconsin-Madison have developed an AI-driven precision medicine tool designed to identify sex-specific risk factors for glioblastoma, according to a recent Science Advances study.
Glioblastoma is the most common primary brain cancer, with an estimated 12,000 new cases diagnosed in the United States each year. Following diagnosis, patients survive an average of 15 to 18 months, and the five-year survival rate for the disease hovers around 10%.
Part of the difficulty in treating the condition lies in the fact that brain tumors, in general, are challenging to treat. But in the case of glioblastoma, tumors form microscopic branches that can spread into different parts of the brain as they grow, making them difficult to remove via surgery.
Treating brain cancers with drugs is also not without hurdles, as these medications often cannot cross the blood-brain barrier. These factors, along with the cancer's aggressive nature, make identifying risk factors key to improving patient outcomes.
The researchers noted that to date, a handful of risk factors for glioblastoma have emerged: the disease develops more often in male, Caucasian patients over the age of 50. High-grade gliomas are typically more aggressive in male patients, as well.
The presence of these disease drivers suggests that other, sex-specific risk factors might influence glioblastoma prognosis. To explore these, the research team turned to deep learning.
"There's a ton of data collected in a cancer patient's journey," explained UW-Madison radiology and biomedical engineering professor Pallavi Tiwari, Ph.D., in a press release. "Right now, unfortunately, it's usually studied in a siloed fashion, and this is where AI has huge potential."
AI has demonstrated significant potential in medical imaging, spurring the researchers to examine whether applying deep learning to hematoxylin and eosin (H&E) scans could help flag sex-specific attributes of a tumor's microenvironment and inform the development of risk profiles for glioblastoma prognosis.
The researchers underscored that these insights could advance risk stratification and precision medicine for glioblastoma patients.
"We want to address the entire spectrum of challenges in a cancer patient's journey, starting from diagnosis and prognosis to treatment response assessment," Tiwari noted.
To do this, the research team trained the model separately on over 250 routine H&E-stained slides from male and female patients with high-grade gliomas. Using this data, the deep learning tool was tasked with identifying unique characteristics of each tumor, like the degree to which it invaded surrounding healthy tissues or the presence and abundance of particular cell types.
From there, the model was trained to flag any patterns that linked these characteristics with patient survival, while accounting for sex.
The resulting analysis highlighted risk factors for aggressive tumors that were strongly associated with sex, such as the presence of pseudopalisading cells in males and degree of tumor invasiveness in females.
"By uncovering these unique patterns, we hope to inspire new avenues for personalized treatment and encourage continued inquiry into the underlying biological differences seen in these tumors," said Ruchika Verma, Ph.D., a computational scientist with the Icahn School of Medicine at Mount Sinai.
Shania Kennedy has been covering news related to health IT and analytics since 2022.
