Laboratory Investigation

Raman spectroscopy (RS) has been shown to differentiate various disease states in tissue samples, yet the impact of sample substrate on spectral quality and downstream machine-learning (ML) performance is under-explored. We collected 960 spectra from 36 formalin-fixed, paraffin-embedded colonic specimens mounted in parallel on calcium fluoride (CaF2) and mirrored stainless-steel (SS) slides. Raw SS spectra were more intense than CaF2, and routine dewaxing left prominent paraffin peaks on CaF2 that were largely absent on SS. Extending CaF2 xylene immersion to 48 h reduced key paraffin peaks by 11 to 23%, aligning its biochemical profile with SS. Nested cross-validated classifiers were built to distinguish true disease (T1) from epithelial misplacement (EM). The highest accuracies were observed on SS (0.65 - 0.66), while CaF2 improved from 0.41 - 0.57 to 0.65 - 0.69 after extended dewaxing, chiefly via gains in recall. Although optimal dewaxing renders CaF2 diagnostically non-inferior, SS still offers superior accuracy, faster preparation, at approximately 4% of the cost, supporting its integration into scalable, AI-enabled RS pathology workflows.

Context: Cytological specimens allow minimally invasive, repeated sampling for diagnosis and molecular testing. However, cytology biobanking protocols have primarily been optimized for morphology, and the effects of preanalytical handling and storage on intracellular proteins remain unclear. Objective: To assess how fixation, staining, storage temperature, and duration affect intracellular protein preservation in cytological specimens and to identify conditions suitable for in situ proximity ligation assay (isPLA). Design: Human cancer cell lines were processed under different preanalytical conditions. Intracellular proteins and protein protein interactions were assessed by isPLA, fluorescence microscopy, and quantitative image analysis. Slides were stored at room temperature or minus 20 Celsius, with or without MayGrunwaldGiemsa (MGG) staining. Results: Storage temperature had the greatest effect on protein signal. Unstained slides stored at minus 20 Celsius retained strong signals for up to 24 months, whereas room temperature storage reduced detectable protein interactions by approximately 95%. MGG staining further reduced signal intensity during prolonged storage. Nevertheless, stained slides remained suitable for isPLA when staining, microscopic evaluation, destaining, and antibody labeling were completed on the same day. Prolonged room temperature storage of stained slides caused loss of antigenicity, increased autofluorescence, and false negative findings. Conclusions: Low temperature storage and standardized preanalytical handling preserve intracellular protein signals in cytological specimens. These conditions support reliable single cell protein profiling of archived samples and their use in biomarker research and translational pathology.

Blood-based metabolomic profiling has been widely investigated for breast cancer (BC) detection; however, clinical implementation remains limited due to variability in sample handling, analytical reproducibility, and overfitting during statistical analysis. We established a plasma GC/MS metabolomics workflow for discriminating BC from healthy controls (HC) using conventional machine-learning algorithms. Plasma samples (n = 360; BC = 180, HC = 180) were collected prospectively under standardized preanalytical conditions before surgery and the initiation of systematic anticancer therapy and analyzed using a quantitative GC/MS platform with automated derivatization. Feature selection and model development were conducted using three machine-learning (ML) algorithms (Lasso logistic regression (LR), random forest classifier (RFC), and support vector machine (SVM)). A total of 45 metabolite candidate biomarkers were identified, and the optimal number of metabolite features for each algorithm was estimated by a recursive feature elimination (RFE)-based strategy. The best-performing models achieved area under the ROC curve values (AUC) of 0.910 (LR), 0.893 (RFC), and 0.843 (SVM). We selected prioritizing candidate biomarkers consistently expressed across the multi-algorithm pipeline. A bagging ensemble model improved stability (AUC = 0.911) and reduced false-positive predictions in the independent HC dataset. In addition, model stability with respect to false-positive predictions was assessed using an independent HC cohort (n = 15) that was collected at a separate institution. These results indicate that a plasma metabolomics workflow combined with conventional multi-algorithm ML, algorithm-specific feature selection, and independent assessment provides stable discrimination between BC and HC in a moderately sized cohort.

Targeted gene panels (TGPs) dominate clinical sequencing, yet most copy-number (CN) signature studies rely on genome-wide assays. Whether signatures can be recovered from TGPs and retain clinical relevance remains unclear. We analyzed two real-world TGP cohorts comprising 1,726 patients across 62 tumor types, including 825 with clinical annotations, and made the underlying data publicly available. TGP-derived signatures recapitulated established biological associations, including links to homologous recombination deficiency and TP53 alterations, concordant with published genome-wide assay-derived CN signatures. Using our detailed clinical data, we found TGP-derived CN signatures associated with overall survival under standard therapies in ovarian, pancreatic, and colorectal cancer. In ovarian cancer, CN2 was associated with CCNE1 amplification and shorter survival under paclitaxel/carboplatin, with the survival association validated in an independent SNP-array cohort. These findings demonstrate that routine TGPs yield biologically and clinically relevant CN signatures with potential as biomarkers for therapy stratification.

Stromal cell composition has been proposed as a potential biomarker in ductal carcinoma in situ (DCIS) for assessing the risk of subsequent recurrence or progression to invasive breast cancer (IBC). However, the characteristics of the tumor microenvironment within DCIS associated with these outcomes remain poorly understood. Here, we perform spatial transcriptomic analyses of 49 primary DCIS specimens from patients with no recurrence, DCIS recurrence, or recurrence as IBC. Cell mapping reveals spatially organized stromal cell gradients within DCIS, with enrichment of TREM2+ macrophages and myofibroblastic cancer-associated fibroblasts (myCAFs) in tumor-proximal niches. Notably, the abundance of TREM2+ macrophages near tumor nests is associated with subsequent recurrence as IBC. Moreover, ligand-receptor pair analysis identifies GRN--SORT1, SPP1--CD44, and LGALS9--CD44 interactions as spatially plausible signaling interactions between TREM2+ macrophages and epithelial cells. Finally, these findings are validated at the protein level by multiplex immunostaining of serial sections. This study characterizes the spatial landscape of the DCIS tumor microenvironment associated with subsequent recurrence, providing a basis for developing biomarkers of disease progression and identifying potential therapeutic targets.

Backgroud Breast cancer is a leading cause of cancer-related mortality, and HR+/HER2-; is the predominant subtype. Molecular pathology, including PIK3CA and ESR1 mutation, informs prognosis and treatment selection. Circulating tumor DNA (ctDNA) offers a non-invasive approach for cancer monitoring, but its clinical utility is limited by tissue plasma discordance, reflecting heterogeneous ctDNA release. The cellular and spatial determinants of this release remain poorly understood, particularly in HR+/HER2-; subtype, where heterogeneity across subclones complicates liquid biopsy interpretation. Results To identify the cellular and spatial determinants of ctDNA release, we developed scMASTER, integrating single-cell transcriptomes and mutations, spatial transcriptomics and matched tumor and plasma DNA sequencing. In ten patients with non-metastatic HR+/HER2- breast cancer, eight recurrent malignant epithelial states showed distinct ctDNA release potential. Antigen-presentation, stress, mitosis and mesenchymal-like states showed high release. PIK3CA and ESR1 mutations were linked to distinct functional states and altered ctDNA detectability. Three spatial architectures, EMT-enriched regions, intraductal luminal progenitors and avascular hypoxic zones, showed high release. Tumor-microenvironment interactions were bidirectional: anti-tumoral immunity correlated with reduced release, whereas immunomodulatory components correlated with increased release. A stratification model based on these features identified high-risk, immunosuppressed patients with poor prognosis, validated across external cohorts. Conclusions Our study establishes a single-cell spatial framework for decoding ctDNA release heterogeneity, linking tumor states, somatic mutations, spatial architecture, and immune interactions to ctDNA release. These findings provide mechanistic insights into tissue-plasma discordance, offer a biological rationale for interpreting ctDNA dynamics, refine patient risk stratification and liquid biopsy-based monitoring in HR+/HER2-; breast cancer.

Background: Patients with ER-positive/HER2-negative early breast cancer of low or intermediate grade are generally considered to have a favourable prognosis, yet a number will recur. This risk is poorly captured by standard clinicopathological factors. AI-based medical devices for whole slide image (WSI) analyses used for risk-stratification often predict on slide-level thereby overlooking spatial localisation of the biomarkers. We hypothesise that the spatial distribution of a previously reported AI-based marker provides independent prognostic value beyond its global score. Patients and Methods: This Swedish multicentre cohort study includes 6106 patients with primary invasive breast cancer: 3078 from the Stockholm region (training) and 3028 from Skane (independent test). Tile-level high-grade morphology scores were generated with ViT_DG, a vision-transformer adaptation of DeepGrade, and the presence of clusters of high-risk tiles localised at the invasive tumour front was defined as a binary biomarker. Primary analyses were pre-specified in the ER-positive/HER2-negative/lymph-node-negative and Grade 1-2 subgroup. Association with breast-cancer recurrence-free interval (BCRFI) and distant metastasis were assessed using multivariable Cox proportional hazards models (HR). Results: In the primary subgroup (n=1536 training; n=1403 test), tumour front clusters were present in 37% and 39% of patients. Cluster-positive status was associated with a 2.30-fold increased risk in the external test. Cluster-negative patients achieved a 10-year BCRFI of 95.3%, equivalent to Grade 1 patients and better than Grade 2 (92.4%), whereas cluster-positive patients (90.2%) did worse than the Grade 3 reference. The association was retained in patients classified as low-risk by ViT_DG (test set multivariable HR 2.86), identifying high-risk patients not captured by the global score. Conclusions: The absence of high-risk morphology clusters at the tumour front is an independent prognostic indicator, identifying a low-risk group independently of the global AI biomarker classification. More broadly, preserving the spatial distribution of AI-derived predictions enables prognostic stratification beyond slide-level aggregation.

Vasculogenic mimicry (VM) describes the ability for tumor cells to mimic endothelial cells and form tumor-lined blood vessels. Although VM is associated with poor prognosis, resistance to therapeutics and metastasis, its presence, relevance and functionality have been debated, largely due to the techniques employed to determine its presence. Dual CD31 and periodic acid-Schiff (PAS) histological staining, the method widely used to detect VM within tumor tissue, is hampered by issues with artifacts of 2D sections and does not assess active blood flow through vessels and hence cannot confirm vessel functionality. Here, we present a 3D imaging pipeline to detect functionally perfused VM and host vessels within mouse tumor models. Mice tumor models are intravenously injected with fluorescently-labelled tomato lectin which binds to the lumen of functional blood vessels. These tumors are thickly sliced, optically-cleared, stained with endothelial markers and imaged using light-sheet microscopy. We apply this imaging modality to image tumors from six breast cancer patient-derived tissue xenograft models, an orthotopic mammary cancer model and two small cell lung cancer circulating-tumor cell explant models, where VM vessels, mosaic vessels and host-VM vessel junctions can be detected. To our knowledge, this is the largest collection of 3D VM vessel images to date, providing convincing evidence of the presence and functionality of VM within tumors and defining the level of evidence necessary to detect VM in vivo.

The prognostic relevance of peritumoral SULF1 transcript signal in oral squamous cell carcinoma (OSCC) is unclear. We assembled a retrospective cohort of 133 patients with 218 primary-site tissue-microarray cores; 131 patients had evaluable boundary features. Multiplex fluorescence imaging combined RNAscope SULF1, CK-pan, and DAPI. Tumor-stroma boundaries were annotated with the SULF1 channel hidden. We screened 50 exported boundary features across cumulative stromal bands of 35-300 um against recurrence-free survival (RFS), overall survival (OS), and disease-specific survival (DSS), with false-discovery-rate correction. The strongest DSS association was signal SD within 150 um (n = 107; 23 disease-specific deaths; HR = 2.20 per SD; 95% CI, 1.64-2.94; q = 2.1 x 10^-6). In 50 repeated 5-fold cross-validations with feature selection repeated within training folds, adding a boundary feature to AJCC stage increased mean held-out DSS c-index from 0.613 to 0.754 (delta C = +0.141; 2.5th-97.5th percentiles across repeats, +0.105 to +0.192). These percentiles describe split variability, not confidence intervals. The 150-um SD feature was selected in 39.2% of DSS training folds. The DSS association persisted after adjustment for mean signal, although the measurements were strongly correlated. Peritumoral SULF1 signal dispersion was associated with outcome beyond an ordinal AJCC-stage model; independent validation is required.

Current state of the art computational pathology foundation models attain satisfying accuracy most general pathology tasks including detection of pancreatic ductal adenocarcinoma (PDAC) but fail to provide sufficient explainability, operating as "black boxes". Some interpretability effort has been made, with models providing attention maps that project back on the slide the models high attended areas that pushed for the given prediction result, without further naming or reasoning explanation which can limit pathologists trust toward the prediction. We developed C3PRO, a cell-centric framework attempting to provide deeper understanding and explainability, by using embeddings extracted using a pathology foundation model, from cell centered tiles containing both the cell and its close neighborhood, which were then clustered into recurrent morphological phenotypes. A spatial graph was additionally constructed on top of the back-projected vocabulary to account for recurrent motifs. The evaluation of the model performance involved a cohort of 809 patients (825 diagnostic biopsies) using nested cross validation. C3PRO was benchmarked against four different multiple instance learning frameworks and achieved on-par performance for diagnostic (0.958 AUC), survival (0.617) and resectability (0.696) prediction, while providing additional insights compared to the strongest baseline models. Each clustered phenotype was probed and extracted as a representative tile mosaic and annotated by two expert pathologists, constituting a phenotype vocabulary available for statistical analysis. These phenotypes were scored by pathologists with physical brightfield microscopy and replicated the independent statistical significance impact on overall survival prediction from diagnostic biopsies.

Pediatric endoscopy is performed only when clinically indicated, limiting access to healthy duodenal tissue. Biopsies with duodenal no pathologic abnormality (NPA) are often used as controls despite the presence of symptoms or inflammatory disease found elsewhere in the gastrointestinal (GI) tract. We characterized pediatric duodenal NPA tissue across clinical, histologic, cellular, and transcriptomic domains. Methods Archival duodenal NPA biopsies were obtained with clinical metadata and hematoxylin-and-eosin whole-slide images (WSIs). Duodenal mRNA-seq data were analyzed from a subset of patients with duodenal NPA. Clinical metadata and WSIs underwent machine-learning analysis, cell populations were quantified from WSIs, and RNA-seq data underwent differential expression and pathway-enrichment analyses. Results The primary cohort included 195 patients with duodenal NPA. Comparisons between patients with non-duodenal GI disease and those with no GI disease showed differences in inflammatory biomarkers and follow-up utilization. Unsupervised clinical clustering identified three clusters with partial enrichment for IBD with colonic inflammation and Eosinophilic Esophagitis (EoE) with esophageal inflammation. Supervised clinical classification showed modest discrimination. WSI clustering showed limited disease-status discrimination, and cell quantification showed no significant group differences. In the separate RNA-seq cohort of 43 patients, differential-expression and pathway-enrichment analyses identified transcriptional and pathway-level differences between disease-status groups. Conclusions This multi-level characterization indicates that pediatric duodenal NPA tissue should not be treated as a uniform control category. Clinical metadata and transcriptomics revealed clinical and molecular heterogeneity, while histologic and cell analyses showed limited disease-status separation, supporting a refined definition of control tissue.

Non-small-cell lung cancer (NSCLC) diagnosis is challenged owing to the need for reliable, non-invasive biomarkers. Circulating cell-free microRNAs (ccf-miRNAs) are promising candidates, but their low levels and short sequences in blood complicate detection. In this study, a novel nano-hybrid fluorescent array was developed to enable rapid and sensitive identification of NSCLC-related ccf-miRNAs. The array consists of poly(T)-modified fluorescent nano-polystyrene beads for capturing target miRNAs, and fluorophore-labeled, sequence-specific locked nucleic acid (LNA) probes for their detection. This design forms a sandwich structure with the target miRNA. Nano cytometry and fluorescence microscopy assessed both the capture and specificity of detection. The assay was tested on plasma samples for miR-16-5p and U6, evaluating selectivity, sensitivity, and reproducibility. The results showed that the poly(T)-modified beads efficiently captured the target miRNAs, and the LNA probes accurately distinguished the sequences. The assay enabled direct detection of miR-16-5p and U6 from plasma without amplification, demonstrating high selectivity, sensitivity, and reproducibility. Combining enrichment with nano-polystyrene beads and sequence-specific LNA probes addressed the main challenges of ccf-miRNA detection, namely their low abundance and short sequence length. As ccf-miRNAs are linked to NSCLC progression, this method could assist in early diagnosis, disease monitoring, and assessment of treatment response. The target-independent nature of the poly(T) capture layer allows for easy calibration to new miRNAs by changing the LNA probes, rendering it suitable for multiplex detection in clinical settings. However, further preclinical and clinical studies are needed before adoption in routine practice. This nano-hybrid fluorescent array presents a rapid and reliable approach for detecting low-abundance ccf-miRNAs in plasma. Its encouraging performance and flexible design show potential for future application in NSCLC diagnosis and monitoring, subject to additional confirmation.

Defining tumour regions on haematoxylin and eosin (H&E) sections is a routine first step in spatial-omics studies, yet it is usually done by hand and is difficult to reproduce. We present OmiCoreTumorDetector, an openly licensed model that maps tumour-enriched regions in colorectal cancer (CRC) H&E sections and exports them as QuPath-compatible annotations. The released model (omicore-tumordetector-crc-he-v0.1) is an ensemble of three convolutional classifiers trained on 100,000 public tissue tiles, combined with Macenko stain normalisation at inference. During development we found that the main obstacle to reuse was calibration under stain-domain shift rather than discrimination: a single model kept an area under the ROC curve (AUROC) of 0.955 on unseen slides while its sensitivity at the conventional 0.5 threshold fell to 0.48. Training on non-normalised tiles raised tumour AUROC on an independently collected tile set from 0.836 to 0.992, and normalising at inference reduced false-positive tumour area in normal-adjacent tissue by 16- to 26-fold. On five 10x Visium HD CRC sections that share no material with the training data, the released model called 27.7-48.5% of tissue as tumour in three carcinoma sections and 0.08% and 1.82% in two normal-adjacent sections, exporting no tumour region from either normal section. On the carcinoma section with matched single-cell-resolution transcriptomics, agreement with transcriptome-derived tumour-cell identities reached an AUROC of 0.985 (95% spatial-block bootstrap CI 0.975-0.993). The image model never observes gene expression, so this is orthogonal evidence. The model localises tumour-enriched regions at 112 um resolution; it does not identify individual malignant cells and has not yet been validated across scanners, institutions or histological variants. Code, weights and evaluation are released under Apache-2.0 and installable with pip install omicoretumordetector.

Background: General-purpose large language models (LLMs) are increasingly evaluated in diagnostic pathology, but prior studies have largely emphasized diagnostic accuracy rather than how models fail. We evaluated four LLMs for diagnostic performance, pathology-relevant errors and hallucinations, their burden, and potential clinical impact across multi-organ pathology cases. Design: In this retrospective multicenter study, 153 pathology cases from two institutions spanning 20 organs were evaluated using ChatGPT-5.3 (LLM1), Gemini 3 (LLM2), Grok 4.20 (LLM3), and Claude Opus 4.6 (LLM4). Each LLM received multi-magnification histologic images with clinical context and generated a microscopic description and diagnosis. No data splitting, training, or fine-tuning was performed. Twenty-one pathologists assessed 612 outputs for diagnostic correctness, error and hallucination type and burden, clinical impact (0-4), and overall performance (1-5). Results: Strict diagnostic accuracy was 48.9% overall (60.1% including partially correct diagnoses) and ranged from 36.6% to 58.2% across LLMs. Errors occurred in 82.0% of outputs and hallucinations in 76.6%; 90.2% contained at least one error or hallucination. Misinterpretation was the most frequent error (72.1%), while fabricated histologic features were the dominant hallucination type (74.5%). LLM2 had significantly lower misinterpretation rates than the other three LLMs, while LLM3 had significantly higher fabricated-feature hallucination rates than LLM1 and LLM2. Errors were more frequent in incorrect than correct diagnoses (97.5% vs 66.2%), as were hallucinations (95.1% vs 59.9%; both p < 0.001). Every strictly incorrect diagnosis contained an error and/or hallucination, while 79.9% of strictly correct diagnoses also contained at least one. In multivariable analysis, LLM3 was independently associated with higher odds of an incorrect diagnosis. Diagnostic correctness was the strongest determinant of high clinical impact; each one-point increase in hallucination burden increased the odds by 63% (adjusted OR, 1.63). Conclusion: Diagnostic accuracy alone substantially underestimates the safety limitations of general-purpose LLMs in histopathology. Errors and hallucinations were common even when the final diagnosis was correct, and their burden was independently associated with clinical impact. Evaluation frameworks should therefore assess both diagnostic correctness and the reliability and potential consequences of accompanying generated content.

Perovskite quantum dots (PQDs) are exceptionally promising next-generation optical materials for bioimaging applications. This paper presents an optical PQDs nanocomposite method to identify bacterial species associated with breast cancer, employing a LSTM deep learning model. The results indicate that the developed PQD-based nano-sensor offers impressive sensitivity, selectivity, and practical application. The four bacterial species identified in the study, Pseudomonas aeruginosa, Vibrio parvula, Acinetobacter baumannii, and Streptococcus vestibularis, were considered the most relevant indicators of breast cancer. With rapid detection, an intuitive design, and improved reliability, the PQD sensor could be an excellent breast cancer diagnostic tool for a point-of-care application. Furthermore, it has the potential to be integrated into healthcare systems, specifically in resource-limited settings, thereby improving the accessibility and efficiency of early breast cancer detection.

Salivary gland adenoid cystic carcinoma (SACC) is a rare, poorly understood malignancy. Here, using spatial transcriptomics, we show that disturbances in epigenetic regulation including chromatin remodeling, epitranscriptomic control by m6A RNA methylation, and RNA splicing may contribute to SACC. We demonstrated that machineries responsible for these processes-SWI/SNF chromatin remodeling complex (CRC), NineTeen Complex (NTC), and m6A RNA methylation complex interact directly. Immunohistochemistry on 48 SACC tumor samples indicated altered abundance of METTL3, METTL14, WTAP (m6A RNA methylation) and PRP19, CDC5L, PLRG1 (NTC-splicing) proteins. Chromatin immunoprecipitation indicated direct regulation of genes encoding subunits of SWI/SNF, NTC and RNA methylation complexes, and some EMT (epithelial-to-mesenchymal) marker genes by BRM (BRAHMA). Transcriptomic re-analysis revealed broad alterations in transcript usage, and splicing events in SACC. Collectively, our findings provide the first evidence that coordinated disruption of chromatin remodeling, RNA methylation, and splicing contributes to SACC development and progression, highlighting new avenues for therapeutic exploration.

Foundation models (FMs) have rapidly become dominant in artificial intelligence and are increasingly being adopted in computational pathology. Numerous pathology-specific FMs have been developed and evaluated for a variety of downstream tasks, most commonly using frozen image embeddings with linear probes. While a pathology FM, at least implicitly suggests broad reusability across tasks, the transferability of these models to neurodegenerative disease-related tasks remains largely unexplored. In this study, we evaluated fourteen frozen feature extractors, including general-vision and pathology FMs, across four pathology image classification datasets spanning two specialized neurodegenerative disease immuno-histochemistry (IHC) tasks (Tau neurofibrillary tangle (NFT) and amyloid-{beta} plaque classification) and two cancer-related hematoxylin and eosin (H&E) tasks (the breast tissue BACH dataset and the multi-tissue TIL dataset). The cancer-related H&E datasets represent tasks more closely aligned with the predominant pretraining domain of many current pathology FMs, whereas the neurodegenerative IHC datasets represent a more specialized domain with limited representation in existing FM pretraining cohorts. Frozen embedding linear probes were compared against a conventional supervised ResNet-50 convolutional neural network (CNN) trained directly on image tiles using identical train, validation, and test splits. Across both neurodegenerative IHC datasets, the supervised CNN substantially outperformed all frozen FM linear probes. In contrast, pathology FMs achieved performance comparable to, and in some cases exceeding, the supervised CNN across the two cancer-related H&E datasets, with UNI2-h achieving the highest performance on BACH and several pathology FMs performing on par with the CNN on the larger TIL dataset. Furthermore, a supervised CNN trained using only 1% of the Tau NFT training data (1,938 tiles) still exceeded the performance of the best frozen FM linear probe trained on the complete dataset. Together, these results suggest the transferability of frozen pathology FMs may depend strongly on how well the downstream task is represented by their pretraining domain. These findings demonstrate frozen pathology FMs transfer effectively to the cancer-related H&E tasks evaluated here but may be less effective for specialized neurodegenerative IHC tasks less well represented in current FM pretraining cohorts. Expanding pathology FM pretraining datasets to include a broader range of disease domains and staining modalities may therefore improve transferability to specialized pathology applications. Our findings also highlight the continued importance of conventional supervised learning and motivate future work investigating nonlinear probes and end-to-end FM fine-tuning.

Cervical cytology screening relies on the visual assessment of nuclear and cellular morphology across thousands of cells per slide, a labor-intensive task that could benefit from quantitative computational support. Here we present CytoLens, a modular deep learning system that detects and counts nuclei, delineates nuclei and whole cells, measures the nuclear-to-cell area ratio (N:C), classifies cells from multifocal images and scores candidate abnormality. We trained and tested each module on a public dataset matched to its task, without pooling datasets. CytoLens detected nuclei with an F1 score of 0.931 (95% confidence interval (CI), 0.914-0.946) and counted 75 of 100 fields to within one nucleus. Point-prompted segmentation achieved mean Dice coefficients of 0.901 for nuclei and 0.847 for whole cells and, without retraining, transferred to a second dataset in which measured N:C increased with lesion grade in parallel with expert outlines. Multifocal hierarchical classification reached a mean accuracy of 0.947 across four Bethesda-aligned categories, and abnormality scores ranked annotated abnormal candidates with an area under the receiver operating characteristic curve (AUROC) of 0.865. These results support the evaluation of CytoLens on patient-linked clinical specimens.

Expansion microscopy (ExM) physically magnifies preserved biological specimens by using a specimen-permeating hydrogel to pull biomolecules, such as proteins, apart from each other. ExM protocols exhibit a tension between the need to initially fix proteins to each other during sample preservation, to retain proteins and their nanoscale organization, and the need to separate proteins from each other isotropically, to achieve physical magnification. We here propose to address this tension by fixing proteins directly to the hydrogel network itself, which then serves as a common substrate for both initial fixation and final separation. We equip proteins with chemical handles, then form a specimen-permeating hydrogel that binds the handles, in a ~15-minute chemical process conducted entirely at -20oC. This network fixation ExM (nfExM) procedure enables excellent epitope retention and nanoscale organization preservation, in cultured cells. nfExM may help support ultrastructural investigations of protein complexes and interactions.

The clonogenic assay is the foundational standard for evaluating cellular radiosensitivity, yet its manual quantification is limited by labor-intensive processing, high inter-observer variability, and a reductive binary scoring system. Here, we present and validate a deep learning-based segmentation pipeline for the automated, high-throughput quantification of mammalian cell survival across diverse radiation qualities (photons, protons, and carbon ions). By benchmarking against expert consensus using Bland-Altman analysis, we demonstrate that this AI framework consistently operates within natural human error margins. Crucially, transitioning to automated whole-well image-based analysis enables the extraction of continuous multidimensional data, including colony area, density, and cellularity. This phenotypic profiling successfully identified and quantified high-cellularity colonies formed by radioresistant BxPC-3 cells under particle irradiation. Coupled with rapid image acquisition platforms, this methodology significantly reduces processing bottlenecks while standardizing the derivation of relative biological effectiveness (RBE). By advancing beyond binary survival metrics to continuous morphological profiling, this framework provides a vital new analytical dimension for investigating structural mechanisms of radioresistance.