Background:Glioblastoma (GBM) is the most common primary malignant brain tumor in adults. While only 5% of GBM cases arise in a familial context, the genetic basis of familial GBM remains unresolved in most affected clusters, suggesting that important susceptibility variants may reside outside recognized cancer-predisposition genes. Proband-based genomic studies of such families provide a rational path to discover rare inherited variants with large biological effects, prioritize candidate genes for functional validation, and define early mechanisms of gliomagenesis that may not be apparent from studies of sporadic tumors alone. Methods:In this study, we investigated a family with GBM clustering spanning two generations. To investigate the possibility of shared germline susceptibility, we enrolled the proband and their two confirmed affected relatives in our study and performed whole-genome sequencing on available whole blood and tumor samples. Rare coding variants shared among the three study participants were identified and the genes harboring these variants were functionally interrogated with pooled loss-of-function CRISPR screens in human neural progenitor cells (NPCs) in vitro and in heterotopic xenograft models. Results:Rare coding variants were identified in 139 candidate genes, and functional genetic screens of human NPCs in heterotopic xenograft models revealed Centrosomal Protein of 126 kDa (CEP126), as the top hit. Genetic disruption of CEP126 conferred a survival and tumorigenic advantage in neural progenitor cells. Conclusion:Together, our results establish a functional framework for interrogating rare cancer germline variants and highlights CEP126 as a biologically tractable GBM predisposition gene for future mechanistic and genetic validation.

Background:Previous machine learning models to intraoperatively predict the molecular status of gliomas using stimulated Raman histology (SRH), such as DeepGlioma, have achieved high performance (91.5% accuracy) on curated datasets. However, when used intraoperatively, DeepGlioma (162M parameters) runs slowly on current SRH hardware and underperforms due to its lack of an image rejection mechanism and its validation on curated images. Here, we introduce SRH-Informed Glioma classificatioN with Attention Learning (SIGNAL) (27M parameters), a lighter model with a built-in attention-based rejection mechanism that outperforms DeepGlioma on uncurated clinical datasets. Methods:SIGNAL was developed using 1.56 million SRH fields-of-view from 967 adult diffuse glioma patients collected between December 2017 and July 2025. We used 412 patients from NYU for training and internal validation and a multi-institutional, international cohort of 555 patients for testing. SIGNAL uses a ResNet50 backbone pretrained using a hierarchical contrastive loss function followed by a multi-head multi-layer perceptron (MLP). Using a patch-based attention threshold of 0.6, a final MLP was trained to predict glioma subtypes: glioblastoma, oligodendroglioma, or astrocytoma. Results:SIGNAL outperformed DeepGlioma, achieving greater overall accuracy (90.10% vs. 72.59%) while running faster (16.0 vs. 6.7 patches/s). SIGNAL also outperformed DeepGlioma on all three molecular classification tasks, including IDH mutation (accuracy: 93.51% vs. 79.22%), 1p19q codeletion (93.51% vs. 88.31%), and ATRX loss (89.61% vs. 83.98%). SIGNAL's attention mechanism had a strong positive linear correlation with mean patch cellularity (r=0.96, p<0.001) and a strong negative correlation with patch blood coverage (r=-0.99,p<0.001). Finally, subtype and molecular accuracy between tumor core and margin samples were equivalent despite significantly lower patch retention in tumor margins (44.5% vs 60.2%, p<0.0001). Conclusion:SIGNAL is a lightweight model for intraoperative molecular classification of gliomas using SRH imaging. Its attention-based image quality filter allows for excellent performance, quick processing, and highly interpretable outputs critical for reliable use in intraoperative workflows. Brief 1-2 Sentence Description:We present SIGNAL, a lightweight machine learning model for intraoperative molecular classification of diffuse gliomas using stimulated Raman histology, whose core innovation is a learned attention mechanism that filters diagnostically uninformative tissue, such as blood and acellular regions, before classification, enabling robust real-world generalizability. Validated on 555 patients across four international centers, SIGNAL outperforms the previous state-of-the-art model DeepGlioma on glioma subtype classification (90.10% vs. 72.59% accuracy) while running 2.4 times faster on intraoperative hardware.

BACKGROUND AND OBJECTIVES:Interactions between cancer cells and their microenvironment are central to tumor formation. Regional microenvironmental variability in the brain may offer insights into essential factors in tumorigenesis. Surprisingly, a granular assessment of regional patterns of gliomagenesis has not been undertaken in the molecular era. The aim of this study was to quantitatively establish the anatomic distribution of the major molecular subtypes of adult diffuse glioma. METHODS:We retrospectively analyzed 204 isocitrate dehydrogenase (IDH)-mutant and 200 IDH-wildtype gliomas. Reproducibility was assessed in an external cohort (190 IDH-mutant, 227 IDH-wildtype), and microarray expressions from Allen Human Brain Atlas were used to compare transcriptomic profiles between IDH-mutant hotspots and coldspots. RESULTS:A total of 50.5% (103/204) of IDH-mutant tumors arose with the superior and middle frontal gyri, indicating a 3.1-fold regional enrichment relative to the volume of these gyri (P < .001). Totally, 9.5% (19/200) of IDH-wildtype tumors arose in the superior temporal gyrus with a 2.1-fold enrichment (P = .01). IDH-mutant and wildtype tumors were enriched by 4 and 4.5-fold, respectively, in the insula (both P < .001). Overall, 23.3% (24/103) of astrocytomas occurred disproportionately higher in the insula compared with oligodendrogliomas (P < .001). Transcriptomic analysis comparing the lobar hotspot (frontal lobe) to the coldspot (occipital lobe) revealed frontal enrichment of cholesterol (normalized enrichment score = 1.78) and fatty acid (normalized enrichment score = 1.94) metabolism pathways, paralleling the observed regional enrichment of IDH-mutant gliomas. CONCLUSION:This study identifies molecular subtype-specific glioma hotspots and may suggest that regional metabolic differences may underlie the brain's variable vulnerability to gliomagenesis. These findings provide a framework for investigating additional microenvironmental factors that drive human glioma formation.

BACKGROUND AND OBJECTIVES:Molecular markers such as isocitrate dehydrogenase (IDH) and alpha-thalassemia/mental retardation syndrome X-linked (ATRX) status are essential for glioma classification and treatment planning, but their manual extraction from pathology reports creates significant research bottlenecks. This study evaluated 3 Natural Language Processing approaches with increasing computational complexity: deterministic Regular Expressions (RegEx), statistical Term Frequency-Inverse Document Frequency (TF-IDF) with logistic regression, and contextual deep learning Bidirectional Encoder Representations from Transformers (BERT). We address whether more intensive approaches provide sufficient performance benefits over simpler approaches in computational pathology research. METHODS:We analyzed pathology reports from 404 patients with glioma at Institution A and 197 at Institution B for external validation. IDH analysis included 399 (Institution A) and 193 (Institution B) patients; ATRX analysis included 361 and 130 patients, respectively. All approaches underwent identical preprocessing steps, including text normalization, terminology standardization, and context extraction. Performance was evaluated using standard classification metrics and memory usage benchmarks on internal and external validation data sets. RESULTS:Simpler approaches outperformed more intensive approaches on external validation. For IDH, Regex achieved near-perfect accuracy (99%, area under the curve [AUC] 1.000) and TF-IDF performed exceptionally (94.2%, AUC 0.984), while BlueBERT underperformed (85.2%, AUC 0.934). For ATRX, Regex achieved perfect accuracy (100%, AUC 1.000) and TF-IDF maintained high accuracy (98.0%, AUC 0.998), outperforming BERT-large (84.6%, AUC 0.931). BERT-based approaches required 1825-1953 MB of memory vs Regex (0.82-5.52 MB) and TF-IDF (17.27-34.89 MB). CONCLUSION:Simple Natural Language Processing approaches effectively automate molecular marker extraction from pathology reports with near-perfect accuracy while requiring minimal computational resources. This enables expanded sample sizes in retrospective studies, multi-institutional analyses of rare molecular subgroups, and accelerated biomarker research. Future work will focus on validation across larger data sets, infrastructure integration, and expansion to additional molecular markers.

Abstract Spinal tumor surgery requires rapid tissue diagnosis to guide surgical decisions and further treatment strategies, yet current intraoperative methods are time-intensive and require specialized expertise. No AI systems exist for real-time spinal tumor classification during surgery. We developed SpineXtract, the first AI-powered system for rapid intraoperative spinal tumor diagnosis using stimulated Raman histology (SRH) — a label-free Raman spectromics imaging technique without tissue processing available during surgery. We created a transformer-based classifier optimized for spinal tissue characteristics to identify common tumor types: meningioma, schwannoma, ependymoma, and metastasis. The system was tested in an international, multicenter, simulated, single-arm study using existing SRH datasets (44 patients, 142 slide-images) from three international institutions, with final pathological diagnosis as reference standard. SpineXtract achieved a 92.9% macro-average balanced accuracy (95% CI: 85.5–98.2) within 5 minutes (tumor-specific accuracy range, 84.2–98.6%), while providing quantitative microscopic feedback for granular tissue analysis. Performance remained consistent across institutions (macro balanced accuracy 91.4–92.0%) and outperformed existing brain tumor classifiers by 15.6%. Our results demonstrate clinical applicability, enabling rapid intraoperative diagnosis with performance exceeding current methods, potentially transforming intraoperative diagnostic workflows in spinal tumor surgery.

Class-defining clonal mutations occur in nearly all cases of adult low-grade gliomas and serve as critical prognostic and diagnostic biomarkers. The ability to unlock tumor mutational status at the time of surgery would significantly elevate the standard of surfgical and adjuvant care for brain tumor patients. However, rapid access to molecular data during surgery is not available, complicating accurate intraoperative diagnosis and detection of tumor cell burden at surgical margins. In 50 specimens collected from 13 adult IDH- or BRAF-mutant glioma patients, we demonstrate the clinical feasibility of intraoperative UR-ddPCR assays for the IDH1 R132H and BRAF V600E clonal mutations. Combined with stimulated Raman histology, UR-ddPCR estimates tumor cell percentage (0-95.0%) and tumor cell burden (<1-1332 cells/mm2) in biopsies taken from tumor core and margins with accuracy comparable to a standard 3-hour ddPCR workflow (R2=0.996, p=5.5x10-57). The UR-ddPCR workflow presented herein represents a 12-fold increase in turnaround time from tissue to molecular detection compared to the current standard of care. With anticipated refinements in assay instrumentation, we anticipate that UR-ddPCR will enable a new standard in point-of-care molecular diagnostics and surgical precision at the cellular level.

Introduction Prostate cancer (PCa) extraprostatic extension (EPE) is common, occurring in ∼30% of men undergoing radical prostatectomy; positive surgical margins (PSM) during radical prostatectomy are frequent as well, occurring in up to 50% of cases with EPE. Stimulated Raman histology (SRH) is a novel microscopic technique allowing real time, label-free, high-resolution microscopic images of unprocessed, un-sectioned tissue providing both morphologic and biochemical information of imaged tissue. We hypothesized that artificial intelligence (AI) utilizing prostate biopsy SRH from MRI identified PCa and clinical variables can predict both EPE and PSM. Methods Prospectively, targeted prostate biopsies from ex-vivo radical prostatectomy specimens, 99% obtained from MRI visible PCa. Prostate biopsies were scanned in a SRH microscope using two Raman shifts: 2845cm-1 and 2930cm-1, to create SRH images. Ex-vivo prostate biopsies were taken from 108 radical prostatectomy specimens to train an attention-based multiple instance learning deep learning neural network (DLNN) made interpretations based on the highest risk patch(s) and/or clinical characteristics as described in Table 1. The DLNN was tested on prostate biopsies from 33 consecutive radical prostatectomy specimens MRI identified PCa (Table 1) yielding receiver operating curve area under the curve (AUC) characteristics. Clinical variables utilized for creation and testing of the DLNN were age, PSA, PSA density, PI-RADS score, region of interest largest dimension and region of interest area. Results The combination of clinical variables and SRH image resulted in an AUC for prediction of PSM of 0.875. Utilizing clinical variables alone, the DLNN predicted PSM with an AUC of 0.787, while assessment of the SRH alone showed an AUC of 0.741.Prediction based on SRH alone, the DLNN predicted EPE with an AUC of 0.795; while analyzing clinical variables alone showed an AUC of 0.59. The addition of clinical variables the SRH DLNN did not further increase the AUC above that of SRH image assessment alone. Conclusions The prediction of EPE alone was best with AI and SRH alone, indicating the clinical factors were not necessary for prediction of EPE. However, the prediction of PSM was best with both clinical and SRH, indicating clinical variables may be required to accurately predict cases where surgeons are likely to perform an incomplete resection. Further validation of these results is required in larger multicentre studies.

Surgical site infections (SSI) can be devastating, yet optimal preparation techniques have not been clearly determined in neurosurgical patients. Furthermore, the profile and origin of pathogens that ultimately go on to cause SSI remain unclear. This is a prospective single-center study including adult patients undergoing elective procedures. Cultures are taken before and after preparation with providone iodine (PVI), chlorhexidine gluconate (CHG), or a combination based on surgeon preference, and again from bone after exposure. Primary outcome is SSI requiring return to the OR. Secondary outcomes are the effects of preparation on surgical site cultures, and the correlation between cultures and ultimate infectious pathogens. The study includes 846 patients undergoing open procedures. 692 cases used CHG, 96 PVI, and 58 both. 11 SSIs required return to the OR with an infection rate of 1.3%. 6 had used CHG (0.9%), 5 PVI (5.2%) during the index surgery. 67 additional patients underwent endoscopic endonasal procedures with PVI prep and no infections. The infection rate for PVI was significantly higher than CHG (p = 0.01), even among matched procedures (p = 0.03). Among the index surgeries with complete surgical site culture data, the causative organism was never present. We found that pathogens present in the first culture but not the second were sometimes present in the third cultures, consistent with the hypothesis that there is a bacterial reservoir in hair follicles inaccessible to skin prep. CHG reduces the rate of SSI. The pathogens that go on to cause infection are not present on the skin at the time of surgery, and thus that causative organisms are introduced into the wound at a later time. This raises important questions about the optimal care of neurosurgical wounds postoperatively.

The CSF-BAM assay, developed by Pearlman, Wang, and colleagues, integrates the detection of somatic mutations, genome aneuploidy, and B- and T-cell receptor clonality from a single cerebrospinal fluid DNA library to increase the sensitivity of cerebrospinal fluid to diagnosis and track brain tumors. See related article by Pearlman et al., p. 2002.

Background In-field or in-margin recurrence after partial gland cryosurgical ablation (PGCA) of prostate cancer (PCa) remains a limitation of the paradigm. Stimulated Raman histology (SRH) is a novel microscopic technique allowing real time, label-free, high-resolution microscopic images of unprocessed, un-sectioned tissue which can be interpreted by humans or artificial intelligence (AI). We evaluated surgical team and AI interpretation of SRH for real-time pathologic feedback in the planning and treatment of PCa with PGCA. Methods About 12 participants underwent prostate mapping biopsies during PGCA of their PCa between January and June 2022. Prostate biopsies were immediately scanned in a SRH microscope at 20 microns depth using 2 Raman shifts to create SRH images which were interpreted by the surgical team intraoperatively to guide PGCA, and retrospectively assessed by AI. The cores were then processed, hematoxylin and eosin stained as per normal pathologic protocols and used for ground truth pathologic assessment. Results Surgical team interpretation of SRH intraoperatively revealed 98.1% accuracy, 100% sensitivity, 97.3% specificity for identification of PCa, while AI showed a 97.9% accuracy, 100% sensitivity and 97.5% specificity for identification of clinically significant PCa. 3 participants’ PGCA treatments were modified after SRH visualized PCa adjacent to an expected MRI predicted tumor margin or at an untreated cryosurgical margin. Conclusion SRH allows for accurate rapid identification of PCa in PB by a surgical team interpretation or AI. PCa tumor mapping and margin assessment during PGCA appears to be feasible and accurate. Further studies evaluating impact on clinical outcomes are warranted.

e14034 Background: The gold standard of diagnosis for central nervous system (CNS) malignancy is biopsy of the tumor, an invasive procedure that, at times, comes with considerable risk or is not feasible. Treatment of CNS malignancies is now heavily based on the molecular profiling of tumor tissue. Without molecular profiling of the tumor, effective treatment decisions cannot be made. We report our development of a pan-CNS cancer liquid biopsy assay, which interrogates cell-free, tumor-derived DNA (cfDNA) found in cerebrospinal fluid (CSF). Our assay uses a novel chemistry, known as TAPS, which enables ultrasensitive and concurrent base sequencing and DNA methylation profiling. This enables detection of the tumor molecular profile from CSF based on sparse tumor-derived cfDNA with minimally invasive collection methods for CSF. Methods: DNA extraction and library preparation from a CSF sample begins with manual isolation of cfDNA from 1-5mL of CSF/sample, processed no more than 48 hours after sample collection. CfDNA is bound to magnetic nanoparticles, washed, and eluted, and cfDNA content is then quantified. Library preparation is performed using TET-assisted pyridine borane sequencing (TAPS chemistry), effectively converting all methylated cytosine (Cs) to thymine (Ts). Samples are then deep sequenced on the NOVASeq 2000.Our bioinformatics workflow leverages a multimodal approach to process and interpret sequencing data derived from TAPS. We align paired-end reads to the hg38 reference genome using BWA-MEM, followed by quality control, to achieve ~10x coverage. Post-processing steps include fragCounter and Dryclean, which implement GC and mappability bias correction and denoising through rPCA, respectively. Methylation signals are extracted using Rastair, a TAPS-specific module enabling genome-wide DNA methylation profiling. Results: TAPS demonstrates robust performance with high sensitivity for detecting CNS tumor biomarkers, enabling concurrent identification of mutations (IDH1/2, TP53, TERT), copy number aberrations (1p19q loss, EGFR amplification, CDKN2A loss), gene fusions (EGFRvIII, MN1:CXXC5), and targeted methylation analysis (including that of the MGMT promoter) within a single assay. Further, principal component analysis of 32 CSF samples showed clear methylation-based classification of CNS tumors, including glioblastomas, meningiomas, and metastatic carcinomas. Conclusions: Using our novel, ultrasensitive pan-CNS cancer CSF liquid biopsy assay, we are able to identify major drivers of CNS disease, both genomic and epigenetic, enabling minimally invasive diagnosis and longitudinal monitoring of CNS cancers. The approach's low DNA input requirements and minimal destructive effects highlight its potential for clinical implementation, particularly for pediatric and metastatic CNS tumors.

Background The diagnosis and treatment of tumors often depend on molecular-genetic data. However, rapid and iterative access to molecular data is not currently feasible during surgery, complicating intraoperative diagnosis and precluding measurement of tumor cell burdens at surgical margins to guide resections. Methods Here, we introduce Ultra-Rapid droplet digital PCR (UR-ddPCR), a technology that achieves the fastest measurement, to date, of mutation burdens in tissue samples, from tissue to result in 15 min. Our workflow substantially reduces the time from tissue biopsy to molecular diagnosis and provides a highly accurate means of quantifying residual tumor infiltration at surgical margins. Findings We demonstrate UR-ddPCR assays for the IDH1 R132H and BRAF V600E clonal mutations that are present in many low-grade gliomas and melanomas, respectively, and whose intraoperative detection would shape surgical decision-making. We illustrate the clinical feasibility of UR-ddPCR by performing it intraoperatively for 22 brain tumor cases, and we further combine UR-ddPCR tumor cell percentage measurements with UR-stimulated Raman histology intraoperatively to estimate tumor cell densities ranging from >1,300 tumor cells/mm2 within a tumor core to <5 tumor cells/mm2 at tumor margins. UR-ddPCR measurements were virtually identical to standard ddPCR measurements performed on the same samples (R2 = 0.995). Conclusions The technology and workflow developed here enable intraoperative molecular-genetic assays with unprecedented speed and sensitivity. We anticipate that our method will facilitate novel point-of-care diagnostics and molecularly guided surgeries that improve clinical outcomes. Funding This study was funded by the National Institutes of Health and NYU Grossman School of Medicine institutional funds. Reagents and instruments were provided in kind by Bio-Rad.

The adoption of large language models (LLMs) in healthcare demands a careful analysis of their potential to spread false medical knowledge. Because LLMs ingest massive volumes of data from the open Internet during training, they are potentially exposed to unverified medical knowledge that may include deliberately planted misinformation. Here, we perform a threat assessment that simulates a data-poisoning attack against The Pile, a popular dataset used for LLM development. We find that replacement of just 0.001% of training tokens with medical misinformation results in harmful models more likely to propagate medical errors. Furthermore, we discover that corrupted models match the performance of their corruption-free counterparts on open-source benchmarks routinely used to evaluate medical LLMs. Using biomedical knowledge graphs to screen medical LLM outputs, we propose a harm mitigation strategy that captures 91.9% of harmful content (F1 = 85.7%). Our algorithm provides a unique method to validate stochastically generated LLM outputs against hard-coded relationships in knowledge graphs. In view of current calls for improved data provenance and transparent LLM development, we hope to raise awareness of emergent risks from LLMs trained indiscriminately on web-scraped data, particularly in healthcare where misinformation can potentially compromise patient safety. Large language models can be manipulated to generate misinformation by poisoning of a very small percentage of the data on which they are trained, but a harm mitigation strategy using biomedical knowledge graphs can offer a method for addressing this vulnerability.

Adult patients (age ≥ 18 years) who have suspected low-grade diffuse glioma. What are the optimal neuropathological techniques to diagnose low-grade diffuse glioma in the adult? Level I Histopathological analysis of a representative surgical sample of the lesion should be used to provide the diagnosis of low-grade diffuse glioma. Level III Both frozen section and cytopathologic/smear evaluation should be used to aid the intra-operative assessment of low-grade diffuse glioma diagnosis. A resection specimen is preferred over a biopsy specimen, to minimize the potential for sampling error issues. Patients with histologically-proven WHO grade II diffuse glioma. In adult patients (age ≥ 18 years) with histologically-proven WHO grade II diffuse glioma, is testing for IDH1 mutation (R132H and/or others) warranted? If so, is there a preferred method? Level II IDH gene mutation assessment, via IDH1 R132H antibody and/or IDH1/2 mutation hotspot sequencing, is highly-specific for low-grade diffuse glioma, and is recommended as an additional test for classification and prognosis. Patients with histologically-proven WHO grade II diffuse glioma. In adult patients (age ≥ 18 years) with histologically-proven WHO grade II diffuse glioma, is testing for 1p/19q loss warranted? If so, is there a preferred method? Level III 1p/19q loss-of-heterozygosity testing, by FISH, array-CGH or PCR, is recommended as an additional test in oligodendroglial cases for prognosis and potential treatment planning. Patients with histologically proven WHO grade II diffuse glioma. In adult patients (age > 18 years) with histologically-proven WHO grade II diffuse glioma, is methyl-guanine methyl-transferase (MGMT) promoter methylation testing warranted? If so, is there a preferred method? There is insufficient evidence to recommend MGMT promoter methylation testing as a routine for low-grade diffuse gliomas. It is recommended that patients be enrolled in properly designed clinical trials to assess the value of this and related markers for this target population. Patients with histologically-proven WHO grade II diffuse glioma. In adult patients (age ≥ 18 years) with histologically proven WHO grade II diffuse glioma, is Ki-67/MIB1 immunohistochemistry warranted? If so, is there a preferred method to quantitate results? Level III Ki67/MIB1 immunohistochemistry is recommended as an option for prognostic assessment. Adult patients (age ≥ 18 years) who have suspected WHO grade II diffuse glioma. Is testing for ATRX mutations helpful for predicting survival and making treatment recommendations? There is insufficient evidence to recommend ATRX mutation testing as a means of predicting survival or making treatment recommendations. Adult patients (age ≥ 18 years) who have suspected WHO grade II diffuse glioma. Does the addition of intraoperative optical histologic methods provide accuracy beyond the use of conventional histologic methods in diagnosis and management? There is insufficient evidence at this time to suggest that intraoperative optical histologic methods offer increased diagnostic accuracy when compared to conventional techniques.

BACKGROUND:Accurate intraoperative diagnosis is crucial for differentiating between primary central nervous system (CNS) lymphoma (PCNSL) and other CNS entities, guiding surgical decision-making, but represents significant challenges due to overlapping histomorphological features, time constraints, and differing treatment strategies. We combined stimulated Raman histology (SRH) with deep learning to address this challenge. METHODS:We imaged unprocessed, label-free tissue samples intraoperatively using a portable Raman scattering microscope, generating virtual H&E-like images within <3 min. We developed a deep learning pipeline called RapidLymphoma based on a self-supervised learning strategy to (1) detect PCNSL, (2) differentiate from other CNS entities, and (3) test the diagnostic performance in a prospective international multicenter cohort and 2 additional independent test cohorts. We trained on 54 000 SRH patch images sourced from surgical resections and stereotactic-guided biopsies, including various CNS neoplastic/nonneoplastic lesions. Training and test data were collected from 4 tertiary international medical centers. The final histopathological diagnosis served as ground truth. RESULTS:In the prospective test cohort of PCNSL and non-PCNSL entities (n = 160), RapidLymphoma achieved an overall balanced accuracy of 97.81% ± 0.91, non-inferior to frozen section analysis in detecting PCNSL (100% vs. 77.77%). The additional test cohorts (n = 420, n = 59) reached balanced accuracy rates of 95.44% ± 0.74 and 95.57% ± 2.47 in differentiating IDH-wildtype diffuse gliomas and various brain metastasis from PCNSL. Visual heatmaps revealed RapidLymphoma's capabilities to detect class-specific histomorphological key features. CONCLUSIONS:RapidLymphoma proves reliable and valid for intraoperative PCNSL detection and differentiation from other CNS entities. It provides visual feedback within 3 min, enabling fast clinical decision-making and subsequent treatment strategy planning.