Large language models (LLMs) and vision-language models represent a fundamentally different category of artificial intelligence (AI) compared to prior image analysis approaches in digital pathology, which have largely been based on convolutional neural network architectures. This review from the American Society of Cytopathology Clinical Practice Committee examines the current evidence for LLM and vision-language model applications in cytopathology, including structured reporting, diagnostic assistance, quality control, education, and workflow integration. The distinction between applications with preliminary evidence and those that remain hypothetical is described. A detailed assessment of the challenges that must be addressed before clinical deployment, including hallucination risk, limited explainability, bias, data privacy, validation gaps, and infrastructure barriers is discussed. A review of the regulatory landscape in the United States and European Union as it applies to AI-enabled software as a medical device is provided. Recommendations addressing cytopathology-specific benchmarks, multi-institutional validation, transparent governance, and incremental deployment beginning with low-risk applications are suggested. In the current environment, LLMs have the potential to augment cytopathology practice, but responsible adoption requires rigorous validation and sustained collaboration among cytopathologists, AI researchers, and regulatory bodies.
BACKGROUND:Deep learning has shown promising performance in cervical cytology; however, many studies have relied on presegmented single-cell images rather than the more complex morphologic patterns encountered in routine practice. Here, scattered cells were defined as isolated or dissociated, nonoverlapping single cells. This study quantified the performance loss when artificial intelligence (AI) models trained on these cells were applied to hyperchromatic crowded cell groups (HCGs). METHODS:Binary convolutional neural network models were developed to differentiate between negative for intraepithelial lesion or malignancy cases and high-grade squamous intraepithelial lesion cases via a scattered cell data set composed of institutional and public liquid-based cytology images. The scattered cell data set comprised 101 cases, with 1062 images; the independent HCG data set comprised 48 cases, with 330 images. ResNet-50, ResNeXt-50, ConvNeXt-Tiny, EfficientNet-B3, VGG-19, and GoogLeNet were trained on scattered cell images, and then directly applied to HCGs without retraining or threshold recalibration. RESULTS:All models showed high performance on the scattered cell data set, with the area under the receiver operating characteristic curve (AUC) ranging from 0.950 to 0.996. When directly applied to HCGs, performance declined across all architectures, with the AUC ranging from 0.385 to 0.683. ConvNeXt-Tiny showed the highest AUC on HCGs (0.683); however, this remained substantially lower than its performance on scattered cells (0.996). For all architectures, the AUC was significantly lower on HCGs than on the scattered cell data set. CONCLUSIONS:Binary AI models trained on scattered cell images achieved excellent discrimination in the original setting but their performance was not preserved when directly applied to HCGs. These findings underscore the need for direct validation and HCG-aware model design in cervical cytology AI.
BACKGROUND:The Lung Cancer Compact Panel (cPANEL) is a recently approved highly sensitive multiplex gene panel in Japan that supports both DNA- and RNA-based next-generation sequencing. Although cytological specimens are acceptable for cPANEL, unfixed cell pellets or dedicated preservation tubes are typically recommended. However, evidence remains limited regarding whether residual liquid-based cytology (LBC) cell suspensions prepared for routine cytological diagnosis can be used directly for cPANEL testing without dedicated molecular preservation or additional preanalytical processing. In this study, we evaluated the feasibility of applying LBC specimens that are widely used in contemporary clinical practice to cPANEL. METHODS:We analyzed DNA and RNA quality in 69 clinical LBC specimens. Among these, 51 specimens containing non-small cell lung cancer cells with previously determined driver alteration status were subjected to cPANEL testing to evaluate assay concordance with clinical companion diagnostic results. RESULTS:DNA integrity was generally well preserved (DNA Integrity Number [DIN]: 6.2 ± 1.5). In contrast, RNA integrity showed greater variability (DV200: 16.4 ± 12.1%). ThinPrep-fixed specimens demonstrated lower DIN and DV200 values compared with CytoRich Red-fixed specimens. Although all samples successfully passed the DNA-based cPANEL assay, six cases (11.8%) failed the RNA-based assay, with RNA yield being a major contributing factor. Among the 46 evaluable specimens, concordance was 95.7% and sensitivity was 92.3%, or 88.9% including RNA module failures as cPANEL-negative. CONCLUSIONS:With appropriate fixative selection and adequate cellularity, cPANEL using clinical LBC specimens may serve as a practical diagnostic platform. We demonstrated that routine LBC specimens can be directly applied to cPANEL without special preanalytical processing.
Two new studies now make the most substantial case yet that the framework of the International System for Reporting Serous Fluid Cytopathology can be carried across the biologic boundary to include cerebrospinal fluid, and current work on the second edition of the system will address cerebrospinal fluid for the first time as a special category of body fluid-an expansion the international community has explicitly endorsed. The contributions of these authors will stand among the empirical foundations on which a coherent, cerebrospinal fluid-specific-and, ultimately, fluid-agnostic-reporting standard can be built.
Recognition of high-risk human papillomavirus (HPV) as the etiologic agent in nearly all cervical cancers has fundamentally reshaped screening strategies, driving the development and adoption of highly sensitive molecular HPV assays. Over the past several decades, cervical cancer screening guidelines have shifted from primarily cytology-based approaches toward HPV-based modalities, with primary HPV testing now recommended as the preferred method by the American Cancer Society and anticipated to be similarly endorsed by the United States Preventive Services Task Force. This article traces the evolution of cervical cancer screening, from the original Papanicolaou test to current guideline recommendations, and discusses the logistical requirements for transitioning to a primary HPV screening system, including triage strategies for HPV-positive cases. The authors additionally present an institutional experience of initiating primary HPV testing, highlighting its impact on cytopathology laboratory workflow and specimen volumes.
This news section is written by a medical journalist and offers Cancer Cytopathology readers timely information on events, issues, and personalities of interest to the subspecialty. In this second of a three‐part series on how new US policies and funding cuts are affecting cancer research, scientists warn that restrictive funding policies may turn off key partnerships.
BACKGROUND:Malignant effusions in breast cancer are a minimally invasive source for molecular profiling. Effusion-derived cytology specimens, cell blocks (CB) and post-ThinPrep PreservCyt (PTPC) supernatant, are reliable sources for molecular testing but differ in DNA concentration and turnaround time (TAT). This study evaluates mutational profiles, cytology characteristics, and clinical correlates to assess the utility of effusion-based molecular testing in breast cancer. METHODS:The authors screened 2400 malignant effusion cases, identifying 22 cases of breast adenocarcinoma effusions that underwent next-generation sequencing (NGS). Both CB and PTPC supernatant were evaluated for tumor cellularity, fluid volume, DNA concentration, TAT, and mutation detection. Associations with clinicopathological features and outcomes were explored. RESULTS:NGS was successful in all 22 cases (mean volume, 426.5 mL; mean tumor cellularity, 50%). Successful sequencing was achieved in low volume samples with high tumor cellularity. Higher tumor cellularity (>50%) was associated with bloody effusions (p = .016). PTPC supernatant yielded 5-fold higher DNA (mean 177.4 ng/µL vs. 33.8 ng/µL) and shorter TAT by 9 days (15 vs. 24 days). TP53 mutations were more common in estrogen receptor (ER)-negative and triple-negative breast effusions (p = .025). Mutations in PIK3CA, TP53, CCND1, and AKT were reliably detected. Time from diagnosis to effusion development was significantly shorter in ER negative disease (p = .001) CONCLUSIONS: Effusion cytology using PTPC supernatant, support successful NGS even in low-volume samples. ER-negative and triple-negative effusions demonstrate more aggressive molecular features including enrichment of TP53 mutations and earlier effusion development. Effusion-based molecular testing yields clinically relevant genomic information in advanced breast cancer.
BACKGROUND:Bladder cancer is a common and highly recurrent malignancy requiring lifelong surveillance. Urine cytology serves as a noninvasive triage tool to guide cystoscopy but is limited by variable sensitivity and manual review. Although deep learning enables quantitative cell-level assessment, limited work has examined three-dimensional urine cytology preparations (e.g., SurePath) containing residual cellular fragments, where restricting analysis to a single nominal focal plane may obscure diagnostically relevant features. This study aimed to quantify focal-plane heterogeneity, measure degradation of nuclear-to-cytoplasmic (NC) ratio and nuclear area estimates off plane, and evaluate focal-plane selection algorithms for performance recovery. METHODS:A total of 325 SurePath whole-slide images scanned as 11-plane Z-stacks were analyzed that spanned negative through high-grade urothelial carcinoma cases. A detection model identified cells and clusters across planes, and 343 clusters (2435 urothelial cells) were reannotated at the optimal nuclear and cytoplasmic focal depths. Classical focus metrics and vision-transformer models were evaluated for focal-plane prediction. A U-Net segmentation model generated NC ratios and nuclear areas, and Spearman correlations compared annotated and predicted measurements across optimal, off-plane, and algorithm-selected conditions. RESULTS:Focal-plane prediction accuracy ranged from 42% to 88%, with classical focus metrics outperforming deep-learning approaches. NC ratio correlation was 0.774 at optimal focus and declined progressively off plane (∼0.50 at ±5 planes). Algorithm-selected planes partially recovered performance (up to 0.748). Similar trends were observed for nuclear area estimation. CONCLUSIONS:Accurate focal-plane selection is critical for artificial intelligence-based assessment of three-dimensional urine cytology. Future work will extend this analysis to cluster- and patient-level outcomes in multi-institutional validation studies.
BACKGROUND:Cerebrospinal fluid (CSF) cytopathology lacks standardized reporting, leading to diagnostic variability and inconsistent clinical management. Although the International System for Reporting Serous Fluid Cytopathology (TIS) has been established for effusions, its utility in CSF remains underexplored. For this study, the authors applied TIS categories to CSF specimens from three academic centers across two continents and evaluated their association with overall survival (OS). METHODS:A retrospective review of pathology databases (2019-2022) across three institutions was performed. Data included demographics, cytologic diagnosis, and OS. Diagnoses were reclassified using TIS criteria. OS was analyzed using Kaplan-Meier curves and Cox proportional hazards models, and multivariable analysis was used for predictors of mortality. RESULTS:In total, 1137 patients underwent CSF cytologic evaluation. Diagnostic distribution was as follows: 3.2% nondiagnostic, 67.1% negative for malignancy, 6.2% atypia of undetermined significance, 2.7% suspicious for malignancy, and 20.8% malignant. Among 579 patients with follow-up, OS differed across categories (p < .001). Mean survival was longest in patients who were diagnosed as negative for malignancy (53.9 months), followed by atypia of undetermined significance (48.4 months), and suspicious for malignancy (40.3 months), and it was shortest in patients who were diagnosed with malignancy (18.7 months). CONCLUSIONS:Leveraging a large, multi-institutional cohort with up to 5 years of follow-up, this study demonstrated that TIS categories effectively stratify CSF specimens into distinct prognostic groups. The authors identified a significant correlation between diagnostic classification and OS, with a clear decline in patient outcomes from those who had diagnoses of negative for malignancy/atypia of undetermined significance to those who had diagnoses of suspicious for malignancy and malignancy. These findings support the TIS system as a standardized, evidence-based framework that enhances risk stratification and clinical decision-making in CSF cytopathology.
BACKGROUND:Accurate peritoneal staging is critical in pancreatic ductal adenocarcinoma (PDAC), as the presence of peritoneal metastases significantly alters prognosis and treatment strategy. In this setting, some centers use peritoneal washings (PWs), which allow the sampling of a large area of the peritoneal surface in the absence of a visible lesion, to improve peritoneal staging and improve selection for curative-intent surgery. METHODS:The authors conducted a retrospective review of 28 patients who underwent cytopathologic evaluation of intraoperative PWs collected between July 2019 to July 2024. RESULTS:There was high concordance between PWs and concomitant biopsy. Six patients had positive PWs. Of these six patients, five of them had concordant findings noted on biopsy. There was one patient who had benign findings on biopsy despite having a malignant PW. In a similar fashion, 22 patients had PWs that were negative. Nineteen of these patients had concordant findings noted on biopsy. The remaining three patients had biopsy-confirmed metastasis despite benign findings noted on PW. CONCLUSION:Peritoneal washings can enhance the detection of peritoneal involvement by PDAC while minimizing false-positive diagnoses.
BACKGROUND:Cerebrospinal fluid (CSF) is an important diagnostic tool for detecting leptomeningeal involvement by malignancy or other etiologies. Currently, there is no standardized reporting system for CSF cytology. This study aims to assess the application of The International System for Serous Fluid Cytopathology (TIS) in categorizing CSF diagnoses and the role of flow cytometry in improving diagnostic performances. METHODS:CSF cases at our institution between January 1, 2023, and December 31, 2023, were reviewed and selected slides were examined. Based on the retrospective review of cytomorphologic findings and flow cytometry results, cases were reclassified into five diagnostic categories according to TIS: nondiagnostic (ND), negative for malignancy (NFM), atypia of uncertain significance (AUS), suspicious for malignancy (SFM), and malignant (MAL). RESULTS:The study cohort included 854 CSF cases from 439 patients. Flow cytometry was performed in 185 (21.7%) cases with a positive result in 23 cases. After applying TIS, final cytological diagnoses included NFM (87.2%), AUS (7.1%), SFM (0.6%), and MAL (5.0%). The calculated risk of malignancy was 0.3% for NFM, 4.5% for AUS, 66.7% for SFM, and 100% for MAL. The diagnostic performance metrics were 95% sensitivity, 100% specificity, and 99% diagnostic accuracy. CONCLUSIONS:The application of TIS to CSF specimens helps standardize reporting terminology, improves diagnostic accuracy, and offers predictive risk stratification. The conjunction with flow cytometry to CSF cytology helps reduce indeterminate diagnosis, especially for those with an uncertain etiology or prior history of hematological malignancy.
The diagnosis of T-cell lymphoma is challenging, and establishing clonality is an important aspect of the workup. Molecular testing to establish T-cell clonality though T-cell receptor (TCR) gene rearrangement polymerase chain reaction is the most common modality and can detect clonal rearrangements in ∼90% of T-cell lymphomas, but is limited by increased turnaround time and cost and the need for sufficient material for testing. In the maturation of T cells, the TCR locus rearrangement will result in the mutually exclusive expression of either the TCR beta-chain constant domain 1 or 2 (TRBC1, TRBC2). Antibodies to TRBC1 and TRBC2 have been used in the setting of flow cytometry, and more recently immunohistochemistry, to establish T cell clonality. This study evaluates the utility of a TRBC1/CD3 dual stain for the diagnosis of T-cell lymphoma. The laboratory information system was searched from 2019 through 2025 for cytology samples with a diagnosis of T-cell lymphoma. Twelve samples were identified along with controls. Dual color immunohistochemistry for TRBC1 and CD3 was performed using a brown chromogen for TRBC1 and a red chromogen for CD3. The proportion of cells staining for TRBC1 and CD3 was scored. A monoclonal T-cell population was detected in all samples of T-cell lymphoma. All B-cell lymphomas and benign samples showed a polyclonal staining pattern consisting of a mixture of brown and red cells. Using immunohistochemistry for TRBC1 paired with CD3 is a reliable surrogate for T-cell clonality testing in cytology specimens and can support a diagnosis of T-cell lymphoma on limited material.
Background Testing patients with non-small cell lung cancer for actionable variants is essential for guiding treatment decisions in accordance with established cancer care guidelines, though limited quantity and quality of tumor tissue often leaves insufficient material for comprehensive testing. Cytopathology specimens obtained through minimally invasive techniques are a potential source of diagnostic material for genomic profiling, though typically challenging to analyze.Methods A total of 85 DNA or total nucleic acid non-small cell lung cancer samples derived from 45 fine-needle aspirate rinse or pleural fluid samples from the Hospital of the University of Pennsylvania archive were tested using the Aspyre Clinical Test for Lung (Tissue) in Biofidelity's CAP/CLIA laboratory. All samples were previously characterized by the Oncomine Precision Assay Genexus assay (the orthogonal reference method).Results Eighty-four of 85 passed Aspyre Lung quality control, one failed. Twenty-six samples were positive for variants in the Aspyre Lung panel: 17 for single nucleotide variants (KRAS, EGFR), three for EGFR insertions/deletions, two for MET exon 14 skipping, and five for gene fusions. Eighty-two of 85 samples were run at standard input levels; three of 85 were run at low input but passed Aspyre Lung controls and include one EGFR exon 20 insertion variant-positive. All results were concordant between methods. Positive Percent Agreement and Negative Percent Agreement were 100%.Conclusions Aspyre Clinical Test for Lung performs effectively on samples derived from fine needle aspirate rinses and pleural fluid. Using these cytology-based specimens for biomarker testing enables pathologists to perform simplified genomic profiling while preserving valuable tissue specimens, potentially reducing the need for additional invasive procedures.
