The NCCN Oncology Research Program (ORP) strives to improve the quality of life for patients and reduce cancer-related deaths by advancing cancer therapies through research. Since the program’s establishment in 1999, the NCCN ORP has brought millions of dollars in research grants to investigators at NCCN Member Institutions. Research grants are provided to NCCN through collaborations with pharmaceutical and biotechnology companies; these grants are in turn used to support scientifically meritorious cancer research efforts.

NCCN ORP studies typically explore new avenues of clinical investigation and seek answers to important cancer-related questions. All studies are approved and funded through a scientific peer-review process and are overseen by the ORP.

This feature highlights an NCCN study funded through the grant mechanism.

Deep Learning Radiogenomic Analysis of Primary Central Nervous System Lymphoma on Brain MRI

Principal Investigator: Andreas Rauschecker, MD, PhD, MSc; 2024 NCCN Foundation^® Young Investigator Award Recipient

Condition: Primary central nervous system lymphoma

Institution: UCSF Helen Diller Family Comprehensive Cancer Center

Primary central nervous system lymphoma (PCNSL) is a disease with a poor prognosis and an increasing annual incidence rate. The overall 5-year survival remains low, but exciting emerging data indicates the prognostic significance of tumor genetics in PCNSL. UCSF researchers recently demonstrated the differential prognoses related to genetic mutations in BTG-1 and ETV-6, and with the emergence of new targeted treatment options for PCNSL, there is a critical need to better understand the relationship between imaging and tumor genetics, such that targeted treatments can be appropriately applied to patients upon noninvasive tumor assessment.

The overarching goal of this study is to relate brain MRI features in patients with PCNSL to tumor genetics, thereby capitalizing on innovative artificial intelligence (AI)–based computer vision methods for automated quantitative brain MRI analysis. The long-term goal is to learn the relationship between specific image features, tumor genetics or other disease subtypes, and disease course, and to harness this understanding for more personalized treatment recommendations. This contribution is expected to provide a reliable, automated quantitative assessment of disease burden in PCNSL, which will also provide the basis for quantitatively and probabilistically relating multiple candidate imaging features to tumor genetics. This is a retrospective study focused on applying a novel technology to a large cohort of patients who were imaged and treated at UCSF for PCNSL.

Aims:

• • Assess validity of fine-tuned deep learning models for assessing tumor burden on brain MRIs in patients with PCNSL

• • Utilize multisequence pretreatment brain MRI for noninvasively predicting multiple genetic mutations, including BTG-1 and ETV-6

Expected Outcomes:

• • Development of a clinically useful AI tool that will identify and volumetrically quantify areas of abnormal FLAIR, T1 postcontrast, and diffusion signal on brain MRI in PCNSL

• • Publication of multiple articles on the radiogenomics of PCNSL

• • Creation of the largest dataset of PCNSL imaging, along with tumor genetics, that currently exists and make the dataset publicly available to other researchers

Contact: Andreas Rauschecker, MD, PhD, MSc • andreas.rauschecker@ucsf.edu

Website: brainlab.ucsf.edu