Chinese Neurosurgical Journal study explores AI tool to predict medulloblastoma subtypes and genetic risk with high accuracy

Medulloblastoma is the most common malignant childhood brain tumor, and outcomes vary widely depending on molecular subtype. Current classification typically requires invasive tissue testing, which can delay risk stratification and treatment decisions. Furthermore, previous studies have not investigated high-risk genetic characteristics with insufficient cohort size. Against this background, a new study was published in Volume 11 of this journal. Chinese Neurosurgical Journal September 15, 2025 We investigated an artificial intelligence (AI) tool to identify subgroups of medulloblastoma based on magnetic resonance imaging (MRI) scans.

The team was led by Dr. Yanon Li from the Department of Radiation Oncology at Capital Medical University in China. The researchers developed a model called MB-CNN that was trained on MRI images of 449 patients treated between 2015 and 2023. The model learns to classify tumors into four major subtypes: wingless (WNT), sonic hedgehog (SHH), group 3, and group 4, which are associated with different outcomes and treatment strategies. The model then classified the tumors well (nearly 8 out of 10 times).

“Our goal is to provide physicians with a rapid, less invasive method to understand their patients’ tumors based on molecular subgroup classification and actionable genetic risk assessment.” says Dr. Lee. But this model went beyond classification. In the second step, we also predicted whether the tumors had specific genetic alterations associated with prognosis, such as TP53 mutations in SHH tumors, group 3 MYC amplification, and group 4 chromosome 11 deletions. The model made these predictions with surprising accuracy. It was 91% for TP53, 84% for MYC, and 87% for chromosome. 11 losses.

“These mutations can tell us a lot about how aggressive the tumor is and how best to treat it.” Dr. Lee explains: “If we can predict this from an MRI, it could make a huge difference for patients.”

To see how the model compared molecular subgroups, the researchers tested the AI ​​against a traditional model using only clinical and radiology data. The old method had an accuracy of about 59%, while the AI ​​had an accuracy of 77.5%. When we combined both approaches into a hybrid model, the accuracy jumped to 82.2%.

This approach may help reduce the time required for molecular risk assessment and provide an additional diagnostic tool in settings where advanced genetic testing is not readily available. However, this study is not without caveats. Because this study was conducted retrospectively at two institutions, scanner variability may have affected performance. Larger prospective multicenter studies will be required to confirm generalizability and clinical utility.

“This is a step towards integrating AI into molecular diagnostics. ” Dr. Lee added:. “This technology has the potential to support more accurate and timely treatment decisions without replacing standard genetic testing.”

About Capital Medical University Beijing Tiantan Hospital

Beijing Tiantan Hospital, affiliated with Capital Medical University, is one of China’s leading hospitals known for its expertise in neurology and neurosurgery. We combine patient care, education, and research to advance medical knowledge while training future health care professionals. The hospital values ​​innovation, teamwork and compassionate care and works closely with national and international partners.

Website: https://m.incsg.com/EN/partner-hospitals/beijingtiantan/

About Dr. Yanong Li of China Capital Medical University

Dr. Yanong Li is a physician at Beijing Tiantan Hospital affiliated with Capital Medical University. Dr. Lee specializes in radiation oncology and is involved in advanced research on brain tumors such as medulloblastoma and intracranial germ cell tumors.

Funding information

This research was supported by the Beijing Natural Science Foundation (L232079), the National Science and Technology Key Project of the Ministry of Science and Technology of China (2022ZD0210100), the National Natural Science Foundation of China (82273343, 82172608, 82101356, and 81902975), and the Beijing National Science Fund for Outstanding Young Scholars. (JQ24040), Beijing Nova Star Program (20220484058), Capital Medical University Outstanding Young Scholar Fund (KCB2304), International Exchange and Cooperation Project (2024-GJJL-10).