Summary: Researchers developed an AI model to identify sex-specific risk factors in aggressive brain cancer, glioblastoma. Using digital pathology slides, the model detects subtle patterns in tumors that correlate with patient survival times and differ by sex.
For instance, tumors in females that infiltrate healthy tissue signal higher risk, while certain cells in male tumors are linked to faster progression. This breakthrough could pave the way for more personalized treatment plans, enhancing the quality of life for glioblastoma patients.
Key Facts
- AI identifies sex-specific risk patterns in aggressive brain tumors.
- Infiltrating tumors in women and specific cells in men correlate with higher risk.
- Findings could lead to personalized, targeted glioblastoma treatments.
Source: University of Wisconsin-Madison
For years, cancer researchers have noticed that more men than women get a lethal form of brain cancer called glioblastoma. They’ve also found that these tumors are often more aggressive in men. But pinpointing the characteristics that might help doctors forecast which tumors are likely to grow more quickly has proven elusive.
University of Wisconsin–Madison researchers are turning to artificial intelligence to reveal those risk factors and how they differ between the sexes.
Radiology and biomedical engineering professor Pallavi Tiwari and her colleagues have published their initial findings in the journal Science Advances, hinting at the promise of AI for improving medical care for cancer patients.
“There’s a ton of data collected in a cancer patient’s journey,” says Tiwari, who is also affiliated with the department of medical physics. “Right now, unfortunately, it’s usually studied in a siloed fashion, and this is where AI has huge potential.”
Few researchers better understand this potential than Tiwari. Arriving at UW–Madison in 2022 to help lead the university’s new AI initiative in medical imaging, Tiwari co-directs the Imaging and Radiation Sciences Program at the Carbone Cancer Center. Her research leverages the computational power of AI models to probe large volumes of medical images and find patterns that could help oncologists and their patients make better-informed decisions.
“We want to address the entire spectrum of challenges in a cancer patient’s journey, starting from diagnosis and prognosis to treatment response assessment,” says Tiwari.
In this case, Tiwari and former graduate student Ruchika Verma turned to digital images of pathology slides — thin slices of tumor samples — in search of patterns that might forecast how quickly a tumor could grow and thus how long a patient might expect to survive.
Glioblastoma is one of the most aggressive forms of cancer, with a median survival of 15 months after diagnosis.
“Patients often don’t have long lives after diagnosis,” says Tiwari. “But a big challenge is prognosis — identifying how long patients are actually going to live and what their outcome is likely to be. This is important because the outcomes ultimately govern the treatments that they’re getting and their quality of life after diagnosis.”
To tackle this challenge, Tiwari and Verma built an AI model that can identify even subtle patterns in pathology slides that might never be apparent to the naked eye. Using data from more than 250 studies of glioblastoma patients, they trained the model to recognize tumors’ unique characteristics, such as the abundance of certain cell types and the degree to which they invade surrounding healthy tissue.
Further, they trained the model to identify any patterns between these characteristics and patients’ survival time while accounting for their sex.
In doing so, they developed an AI model that was able to identify risk factors for more aggressive tumors that are strongly associated with each sex. For females, higher-risk characteristics included tumors that were infiltrating into healthy tissue. Among males, the presence of certain cells that surround dying tissue (called pseudopalisading cells) was associated with more aggressive tumors.
The model also identified tumor characteristics that appear to translate to worse prognoses for both men and women.
The study could help lead to more individualized care for glioblastoma patients.
“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,” Verma says.
Tiwari and her colleagues are doing similar work using MRI data and have begun using AI to analyze pancreatic and breast cancers with the aim of improving outcomes for patients.
In addition to her research, Tiwari is helping to shape the university’s RISE-AI and RISE-THRIVE initiatives, which are establishing UW–Madison as a leader of cross-disciplinary research on artificial intelligence and the human health span, respectively.
“UW has a rich and diverse expertise across our engineering and medical campuses,” says Tiwari, “and with the RISE initiatives, we are well positioned to be at the forefront of translating AI research in clinical care.”
About this AI and brain cancer research news
Author: Will Cushman
Source: University of Wisconsin-Madison
Contact: Will Cushman – University of Wisconsin-Madison
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Sexually dimorphic computational histopathological signatures prognostic of overall survival in high-grade gliomas via deep learning” by Pallavi Tiwari et al. Science Advances
Abstract
Sexually dimorphic computational histopathological signatures prognostic of overall survival in high-grade gliomas via deep learning
High-grade glioma (HGG) is an aggressive brain tumor. Sex is an important factor that differentially affects survival outcomes in HGG.
We used an end-to-end deep learning approach on hematoxylin and eosin (H&E) scans to (i) identify sex-specific histopathological attributes of the tumor microenvironment (TME), and (ii) create sex-specific risk profiles to prognosticate overall survival.
Surgically resected H&E-stained tissue slides were analyzed in a two-stage approach using ResNet18 deep learning models, first, to segment the viable tumor regions and second, to build sex-specific prognostic models for prediction of overall survival.
Our mResNet-Cox model yielded C-index (0.696, 0.736, 0.731, and 0.729) for the female cohort and C-index (0.729, 0.738, 0.724, and 0.696) for the male cohort across training and three independent validation cohorts, respectively.
End-to-end deep learning approaches using routine H&E-stained slides, trained separately on male and female patients with HGG, may allow for identifying sex-specific histopathological attributes of the TME associated with survival and, ultimately, build patient-centric prognostic risk assessment models.
