Molecular Inference–Based Deep Learning Assesses CNS Tumor Diagnosis
The accuracy of a hierarchical molecular inference–based deep-learning system, Neuropath-AI, was tested in central nervous system (CNS) tumor diagnosis and classifications in a retrospective study. Findings from the study were published in The Lancet Oncology.
“Our model provides the basis for a clinically applicable deep-learning assistant to improve human efficiency and accuracy of CNS tumor diagnosis,” H. Lalchungnunga, PhD, of the Laboratory of Pathology, National Cancer Institute, and other study authors wrote in their published report.
Model Methods
The multi-institutional study included data from whole slide images of samples from patients (aged 0–95 years) diagnosed with primary or recurrent CNS tumors. Reference diagnostic labels were determined using DNA methylation-based tumor classification to match 1 of 52 tumor types representing most types within classes of gliomas, embryonal tumors, and meningeal and mesenchymal tumors found in clinical practice.
The Neuropath-AI model was trained using 5,835 samples from the U.S. National Cancer Institute (NCI), U.S. Children’s Brain Tumor Network, and Austrian Digital Brain Tumour Atlas to infer molecular features that were used to predict tumor types. The test cohort consisted of 5,516 samples identified between May 2024 and May 2025 from the National Cancer Institute, U.S. Northwestern Medicine, University of Pittsburgh Medical Center, and University College London. The primary objective was to assess the classification accuracy of the model family–level and terminal classification predictions in test samples.
Key Findings
In the test cohort, tumor family–level classifications were obtained in 5,299 (96%) of 5,516 samples. Predictions of placement in one of the terminal classifications were obtained for 4,772 (87%) samples with at least moderate confidence.
The single highest-scoring classification matched the reference label in 3,817 (95% confidence interval [CI] = 3,770–3,865; 80%, 95% CI = 79%–81%) of 4,772 samples (balanced accuracy [accuracy across all classes] = 66%, 95% CI = 63%–70%). The two highest-scoring classifications contained the reference label in 4,103 (95% CI = 4,056–4,152; 86%, 95% CI = 85%–87%) of 4,772 samples (balanced accuracy = 75%, 95% CI = 71%–78%).
The investigators noted: “The model will be made publicly available and could be implemented to assist human pathologists in future prospective studies.”
Kenneth Aldape, MD, of the Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, Minnesota, is the corresponding author for The Lancet Oncology article.
DISCLOSURES: The study was funded by The Intramural Research Program of the National Institutes of Health. For full disclosures of the study authors, visit thelancet.com.
ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.
Performance of a convolutional neural network in determining differentiation levels of cutaneous squamous cell carcinomas was on par with that of experienced dermatologists, according to the results of a recent study published in JAAD International.
“This type of cancer, which is a result of mutations of the most common cell type in the top layer of the skin, is strongly linked to accumulated [ultraviolet] radiation over time. It develops in sun-exposed areas, often on skin already showing signs of sun damage, with rough scaly patches, uneven pigmentation, and decreased elasticity,” stated lead researcher Sam Polesie, MD, PhD, Associate Professor of Dermatology and Venereology at the University of Gothenburg and Practicing Dermatologist at Sahlgrenska University Hospital, both in Gothenburg, Sweden.