Back to Top
Charles H. Hood Foundation | Volker Hovestadt, Ph.D. – July 2021
By identifying innovative pediatric advancements and providing funding in the critical phases of development, we are able to expedite high-impact breakthroughs that improve the health and lives of millions.
portfolio_page-template-default,single,single-portfolio_page,postid-18316,ajax_fade,page_not_loaded,,qode-title-hidden,qode-child-theme-ver-1.0.0,qode-theme-ver-10.0,wpb-js-composer js-comp-ver-4.12,vc_responsive

Volker Hovestadt, Ph.D.

Assistant Professor of Pediatrics

Dana-Farber Cancer Institute

Machine Learning-based Risk Prediction for Children with Medulloblastoma


Key Words: Medulloblastoma, Childhood brain tumors, Molecular diagnostics, DNA methylation, Risk prediction, Machine learning/Artificial intelligence

Brain tumors are among the most devastating medical conditions in humans. While brain tumors are most common in adults, there are certain types that only affect children. In fact, among all cancers that occur in children, brain tumors are now responsible for the most deaths. Surviving patients often develop long-term neurological and cognitive deficits. This is due to brain injury caused by the tumor itself and treatment with chemotherapy and radiation therapy, which can have detrimental effects on the developing nervous system. Treatment intensity therefore needs to carefully balance the risk of not eliminating the tumor with the adverse effects caused by the treatment itself.

My past and future research centers around better understanding, diagnosing and treating brain tumors in children. There is great potential in using modern genomic techniques and advanced computer algorithms (artificial intelligence) to make accurate predictions if a patient is at high or low risk of relapsing and dying of their disease. These predictions can then be used to guide decisions on treatment intensity for young patients.  Currently, these predictions are based on a small number of clinical and molecular parameters. I propose to develop a model that utilizes half a million of epigenetic markers that are measured in tumor tissues from patients. (“Epigenetic” refers to modifications of the genome that determine how it is regulated). To create this prediction model, my team and I will use data from more than one thousand patients that have been enrolled in clinical trials. Our machine learning algorithms will then automatically “learn” complicated combinations of epigenetic markers that provide accurate predictions in a much better way than any human could do.

The resulting prediction model will be a valuable tool in research and clinical practice, with the potential to improve treatments and lives of children with brain tumors. The concepts and technologies developed in this project will function as an early example of the application of artificial intelligence in healthcare and could be broadly applied to other types of cancer.