Innovative AI Technique Tailored for Predicting Brain Metastases

The best option for predicting brain cancer outcomes perhaps lies in using AI Eye

A novel artificial intelligence (AI) method that York University researchers have created appears to be significantly more effective than the human eye at forecasting treatment results in patients with brain metastases, according to a recent study. The research team is optimistic that fresh developments in technology and research will eventually result in more individualized cancer treatment regimens and improved patient outcomes. The study's principal investigator, York Research Chair Ali Sadeghi-Naini, an associate professor of biomedical engineering and computer science at the Lassonde School of Engineering, calls it "a sophisticated and thorough analysis of MRIs to find features and patterns that are not usually captured by the human eye."

In a case where time is of the essence, we expect that our method, which is a unique AI-based prediction way of detecting radiation failure in brain metastasis, would be able to assist oncologists and patients in making better-informed decisions and adjusting treatment. Previous research has shown that oncologists can accurately predict treatment failure (defined as continued tumor growth) about 66% of the time using standard procedures like MRI imaging, which allows them to determine the size, location, and number of brain metastases as well as the primary cancer type and patient's condition. The top AI model developed and tested by the researchers had an accuracy rate of 83%. When initial malignancies in the lungs, breasts, colon, or other regions of the body travel to the brain through the circulation or lymphatic system, brain metastases, a form of malignant tumor, arise. While there are many different treatment methods, stereotactic radiotherapy is one of the most popular ones and involves administering high-dose radiation directly to the tumor. Not all tumors respond to radiation therapy; up to 30% of these individuals still experience tumor growth even after treatment, according to Sadeghi-Naini. This is frequently not identified until several months following treatment through follow-up MRI.

Patients with brain metastases cannot afford this delay because it is a very crippling condition and most sufferers pass away between three months and five years following diagnosis. Sadeghi-Naini says, "Even before therapy begins, it's crucial to forecast the therapy response. The researchers built artificial neural networks educated on a vast amount of data using a machine-learning method known as deep learning and then trained the AI to pay more attention to particular locations. According to Sadeghi-Naini, when viewing an MRI, you might notice regions inside or around a tumor where the intensity and pattern are different. As a result, your visual system pays more attention to those regions. "However, an AI algorithm is unaware of this. These AI tools can learn which portion of these photos is more relevant and give more weight for analysis and prediction, thanks to the attention mechanism we implemented into the algorithm.

The study was published in the IEEE Journal of Translational Engineering in Health and Medicine and is now accessible online. Ali Jalalifar, a York Ph.D. student and the study's first author, worked in Sadeghi-lab Naini's at York's Keele Campus to complete the modeling work, which was partially financed by the Terry Fox Research Institute (TFRI). The team was able to take advantage of York's long-standing working relationship with Sunnybrook Health Sciences Centre in Toronto when it came to data gathering and evaluating the outcomes from more than 120 patients. The Hatch Memorial Foundation and the Natural Sciences and Engineering Research Council of Canada (NSERC) were additional funders of the project. While further research is required, according to Sadeghi-Naini, the results suggest that AI may one day play a crucial role in the precision control of brain metastases and even other cancer types. Looking at a bigger cohort with a multi-institutional data set would be the next step in implementing this as a therapeutic practice; from there, a clinical trial might be conducted. "There's a fair probability that the overall survival of the patients can be increased if routine therapies can be adjusted for patients depending on their reaction to treatments – that can be anticipated before therapy even starts," he says.