Artificial intelligence (AI) models may allow earlier noninvasive identification of pathogenic variants in central nervous system (CNS) tumors, according to researchers presenting new data at the Association for Molecular Pathology (AMP) annual meeting. By integrating genetic and imaging data, the models could help clinicians anticipate tumor biology and guide treatment before surgical tissue is available.
The study, conducted by a team from Soonchunhyang University in South Korea, was presented Saturday by Jieun Kim, MD, PhD, at AMP 2025 in Boston. The researchers developed 2 machine-learning models: a dense neural network trained on mutation data from 12 key genes using next-generation sequencing of cerebrospinal fluid (CSF)–derived circulating tumor DNA, and a convolutional neural network trained on standardized magnetic resonance imaging (MRI) images. Data were drawn from paired tissue, CSF, and MRI data sets from patients with CNS tumors. The researchers used averaging or majority voting to combine the models’ outputs, optimizing diagnostic robustness.
The models classified samples into 3 categories—glial tumors (GBM), nonglial CNS tumors, and normal tissue—with strong reliability. Misclassifications occurred mostly between glial and nonglial tumors, whereas normal samples were consistently identified. Each model independently achieved high classification accuracy—the dense neural network showed a Matthews Correlation Coefficient (MCC) of 0.8822, whereas the MRI-based convolutional network achieved an MCC of 0.8525. Combining the outputs further improved performance, showing high consistency in identifying pathogenic variants and tumor type.
The findings suggest that multimodal AI may enhance precision neuro-oncology by offering a noninvasive, preoperative tool for mutation detection and histologic classification. Integration of genetic and imaging features could shift clinical workflows toward earlier intervention and personalized treatment planning, reducing reliance on repeated, invasive biopsies, Dr. Kim said.
