Histopathologists Research Articles

Generative Models in Medical Visual Question Answering: A Survey

Medical Visual Question Answering (MedVQA) is a crucial intersection of artificial intelligence and healthcare. It enables systems to interpret medical images—such as X-rays, MRIs, and pathology slides—and respond to clinical queries. Early approaches primarily relied on discriminative models, which select answers from predefined candidates. However, these methods struggle to effectively address open-ended, domain-specific, or complex queries. Recent advancements have shifted the focus toward generative models, leveraging autoregressive decoders, large language models (LLMs), and multimodal large language models (MLLMs) to generate more nuanced and free-form answers. This review comprehensively examines the paradigm shift from discriminative to generative systems, examining generative MedVQA works on their model architectures and training process, summarizing evaluation benchmarks and metrics, highlighting key advances and techniques that propels the development of generative MedVQA, such as concept alignment, instruction tuning, and parameter-efficient fine-tuning (PEFT), alongside strategies for data augmentation and automated dataset creation. Finally, we propose future directions to enhance clinical reasoning and intepretability, build robust evaluation benchmarks and metrics, and employ scalable training strategies and deployment solutions. By analyzing the strengths and limitations of existing generative MedVQA approaches, we aim to provide valuable insights for researchers and practitioners working in this domain.

Application of Electronic Pathology Reading Library in Standardized Practical Skill Training of Clinical Pathology Residents

Objective: To enhance the reading skills of clinical pathology residents, it is essential to establish a well-structured electronic pathology reading library. Methods: In accordance with the Resident Standardization Training Content and Standards (2022 Edition), clinical pathology residents are required to master pathological diagnoses across 11 systems: skin, head and neck, mediastinum and respiratory, digestive, urinary and male reproductive, female reproductive and breast, lymphatic and hematopoietic, bone and soft tissue, cardiovascular, central nervous, and endocrine diseases. Senior pathologists specializing in each subspecialty selected classic pathological slides, which were systematically scanned and compiled into an electronic pathology library. Results: A questionnaire survey was conducted to gather feedback on the electronic pathology reading library. Residents generally found it to be convenient, efficient, and conducive to learning. Conclusion: Training in clinical pathology diagnosis is a core component of standardized resident training. The electronic pathology reading library has been well-received and recognized by resident doctors. However, further efforts are needed to explore diverse teaching methods that align with modern educational approaches, ultimately contributing to the development of highly skilled resident doctors.

OCDet: A comprehensive ovarian cell detection model with channel attention on immunohistochemical and morphological pathology images.

OCDet: A comprehensive ovarian cell detection model with channel attention on immunohistochemical and morphological pathology images.

Glo-net: A dual task branch based neural network for multi-class glomeruli segmentation.

Glo-net: A dual task branch based neural network for multi-class glomeruli segmentation.

A Multimodal Deep Learning Approach for Pre-Operative Tumor-Node-Metastasis Staging in Oral Cancer Using Computed Tomography Imaging and Pathological Slide Data

Oral cancer is one of the most prevalent cancers globally, characterized by high rates of recurrence and metastasis, making early diagnosis crucial for improving patient survival. Accurate preoperative tumor-node-metastasis (TNM) staging is essential for determining effective treatment plans and surgical strategies. Although pathological examination remains the gold standard for TNM staging of oral cancer, imaging data can complement pathology, helping clinicians make more precise preoperative assessments and addressing the limitations of single-modality approaches. In this study, we propose a novel multimodal model for oral cancer staging that integrates preoperative CT images and whole-slide imaging (WSI). To overcome the heterogeneity between CT and WSI features, especially the lack of interaction between macrolevel CT features and microlevel pathological features, we introduce a CT-guided collaborative attention module. Specifically, CT features serve as queries, while patches of WSI are treated as key-value pairs. The corresponding keys are computed through a fully connected layer with trainable weights. The model architecture includes two separate pathways for feature extraction: CT features are extracted using a U-Net-based network, while WSI features are extracted with a multi-instance network utilizing a hierarchical attention mechanism. The CT-guided collaborative attention module facilitates interaction and fusion between these two modalities, resulting in a unified feature representation. This fused feature is then passed through a fully connected layer to produce the final TNM staging prediction for oral cancer. By combining both macro- and microlevel features, our model addresses the limitations of traditional single-modality methods, enabling more accurate tumor boundary delineation. Compared to existing techniques, our multimodal approach improves diagnostic accuracy and provides more reliable staging results. This advancement has the potential to significantly enhance early diagnosis and treatment strategies for oral cancer, offering a more comprehensive and precise method for staging the disease.

Redefining phenotypic intratumor heterogeneity of pancreatic ductal adenocarcinoma: a bottom-up approach.

Pancreatic ductal adenocarcinoma (PDAC) tumor interpatient heterogeneity has been well described with two major prognostic subtypes (classical and basal-like). An important intrapatient heterogeneity has been reported but has not yet been studied extensively due to the lack of standardized, reproducible, and easily accessible high-throughput methods. We built an immunohistochemical (IHC) tool capable of differentiating RNA-defined classical and basal-like tumors by selecting relevant antibodies using a multistep process. The successive stages of (i) an in silico selection from a literature review and a bulk transcriptome analysis of 309 PDACs, (ii) a tumor-specific selection from 30 patient-derived xenografts and single-cell data, followed by (iii) the validation on tissue microarrays in 50 PDAC were conducted. We used our final IHC panel on two independent cohorts of resected PDAC (n = 95, whole-slide, n = 148, tissue microarrays) for external validation. After digitization and registration of pathology slides, we performed a tile-based analysis in tumor areas to identify relevant marker combinations. Sequential marker selection led to the following panel: GATA6, CLDN18, TFF1, MUC16, S100A2, KRT17, PanBasal. Four different phenotypes were identified: one classical, one intermediate (KRT17+), and two basal-like (MUC16+ versus S100A2+) with specific biological properties. The presence of a minor basal contingent drastically reduced overall survival [hazard ratio (HR) = 1.90, p = 0.03], even in classical predominant PDACs. Analysis of preneoplastic lesions suggested that pancreatic carcinogenesis might follow a progressive evolution from classical toward a basal through an early intermediate phenotype. In conclusion, our IHC panel redefined and easily assessed the high degree of intra- and intertumoral heterogeneity of PDAC. © 2025 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Association of a novel AI-based digital pathology (CHAI) biomarker with progression free survival (PFS) in patients (pts) with metastatic hormone sensitive prostate cancer (mHSPC).

204 Background: In mHSPC, no baseline biomarker reliably predicts progression-free survival (PFS). Identifying pts unlikely to benefit from current therapies could enable personalized management and better pt selection for clinical trials. Herein, we investigate the capacity of a novel AI-based digital pathology platform to prognosticate in mHSPC. Methods: Pts diagnosed with mHSPC with available pathology slides and undergoing systemic therapy at the Huntsman Cancer Institute, University of Utah were eligible. Whole slide images were generated from digital scans of diagnostic H&E histopathology specimens reviewed by board-certified genitourinary pathologists. The CHAI (Correlative Histologic AI) platform was used to segment nuclei and extract morphologic and spatial features. Association with PFS was assessed across extracted features, and a biomarker signature was generated. Kaplan Meier estimators, log-rank tests and t-tests were used to compare outcomes among those with and without the biomarker signature for time-to-event and binarized endpoints respectively. PFS was defined from the start of therapy for mHSPC to biochemical or radiographic progression (per PCWG2) or death. Multivariate analysis for PFS was conducted using the Cox proportional hazards model, evaluating the biomarker signature and adjusting for age, Gleason score, baseline PSA, de novo status, disease volume, and ADT intensification. Results: 86 pts were eligible and included. Median age: 65 years (IQR 60-72), median baseline PSA: 29.9 ng/ml (IQR 7.2-23.2) and 64% pts received ADT intensification. 43 pts were categorized as biomarker positive (+) and 43 as biomarker negative (-). Biomarker + pts had a significantly longer median PFS of 67.5 months as compared to 26.9 months in biomarker – pts (HR 0.39, 95% CI HR 0.29 - 0.52, p < 0.001). Biomarker + group was significantly associated with PSA nadir < 0.2 ng/ml (p < 0.001). On multivariate analysis, biomarker + signature was significantly associated with better PFS (Table). Conclusions: CHAI, a novel AI-based digital pathology biomarker is independently associated with PFS in mHSPC setting. After external validation it may serve as a prognostic tool to guide clinical decision-making. Further analysis with a larger sample size will be conducted to evaluate the association with OS. This analysis exemplifies the power of integrating artificial intelligence into medical practice, with a potential to improve outcomes in our patients. Characteristic PFS HR (95% CI) p value Biomarker (+ vs. -) 0.47 (0.24 – 0.91) 0.025 ADT intensification (Yes vs. No) 0.3 (0.16 – 0.7) <0.001 Disease Volume (Low vs. High) 0.37 (0.19 – 0.7) 0.003 Gleason score 1.56 (1.07 – 2.27) 0.021 Age 0.97 (0.93 – 1.00) 0.075 Baseline PSA 1.00 (1.00 – 1.00) 0.061 De novo (Yes vs. No) 1.1 (0.53 – 2.29) 0.793

Predicting long term outcomes following radical prostatectomy using a validated pathology-based multimodal artificial intelligence biomarker.

364 Background: Over the past decade, multiple biomarkers have been developed and validated to improve precision of prostate cancer prognosis estimation. However, these require access to archival tissue and central laboratory processing for RNA extraction, and as a result are labor-intensive and expensive. Deep learning models based on image analysis of standard clinical pathology slides may offer similar accuracy with much improved access and lower cost. Multimodal artificial intelligence (MMAI) models (Artera, Los Altos, CA) have previously been validated to prognosticate as well as predict response to intensified treatment among patients undergoing radiation therapy and have been endorsed in the NCCN prostate cancer guidelines. We determined the accuracy of an MMAI model in prognosticating outcomes after radical prostatectomy (RP) using tissue microarray (TMA) specimens. Methods: We previously built a TMA from 424 mostly low- to intermediate-risk RP cases with rich clinical annotation and long-term follow-up. We scanned this TMA at high resolution and determined MMAI (scored 0 to 1) and Cancer of the Prostate Risk Assessment post-Surgical (CAPRA-S, scored 0-12) scores. We performed logistic regression to determine the MMAI’s score association with adverse pathology (pT≥3a and/or grade group ≥3), and Cox proportional hazards regression to determine the independent ability of the MMAI score to predict biochemical recurrence (two PSA tests ≥0.2 ng/ml or any second treatment) and metastasis after RP. Odds ratios (OR) and hazard ratios (HR) were determined per 0.1 increase in MMAI score. Results: The TMA was successfully segmented and images and MMAI scores were able to be generated for 414 (98%) of cases. Median (IQR) follow-up was 13.2 (7.8-18.3) years. By CAPRA-S, 273 (66%), 114 (28%), and 24 (6%) were low (0-2), intermediate (3-5), and high (≥6) risk, respectively. Recurrence-free and metastasis-free survival were 74% and 96% at 10 years, respectively. Median (IQR) MMAI risk scores were 0.25 (0.18-0.33). On logistic regression, the MMAI score was significantly associated with risk of adverse pathology (OR: 1.05, 95% CI 1.03-1.07, p<0.01). On Cox regression adjusting for CAPRA-S, MMAI risk score was significantly associated with risk of both recurrence (HR 1.04, 95% CI 1.02-1.06, p<0.01) and metastasis (HR 1.05, 95% CI 1.02-1.07, p<0.01). Conclusions: In this cohort of largely lower-risk RP patients, the MMAI score derived from RP TMA samples —originally developed among radiation therapy patients on prostate biopsy samples—adds independent prognostic information above a well-validated multivariable clinical risk model. The MMAI platform allows for rapid, non-destructive analysis of standard pathology slides and should compare favorably to RNA-based platforms in terms of access, speed, and cost.

Assessing Genotype-Phenotype Correlations with Deep Learning in Colorectal Cancer: A Multi-Centric Study.

Deep Learning (DL) has emerged as a powerful tool to predict genetic biomarkers directly from digitized Hematoxylin and Eosin (H&E) slides in colorectal cancer (CRC). However, few studies have systematically investigated the predictability of biomarkers beyond routinely available alterations such as microsatellite instability (MSI), and BRAF and KRAS mutations. Our primary dataset comprised H&E slides of CRC tumors across five cohorts totaling 1,376 patients who underwent comprehensive panel sequencing, with an additional 536 patients from two public datasets for validation. We developed a DL model using a single transformer model to predict multiple genetic alterations directly from the slides. The model's performance was compared against conventional single-target models, and potential confounders were analyzed. The multi-target model was able to predict numerous biomarkers from pathology slides, matching and partly exceeding single-target transformers. The Area Under the Receiver Operating Characteristic curve (AUROC, mean ± std) on the primary external validation cohorts was: BRAF (0·78 ± 0·01), hypermutation (0·88 ± 0·01), MSI (0·93 ± 0·01), RNF43 (0·86 ± 0·01); this biomarker predictability was mirrored across metrics and co-occurrence analyses. However, biomarkers with high AUROCs largely correlated with MSI, with model predictions depending considerably on MSI-associated morphology upon pathological examination. Our study demonstrates that multi-target transformers can predict the biomarker status for numerous genetic alterations in CRC directly from H&E slides. However, their pre-dictability is mainly associated with MSI phenotype, despite indications of slight biomarker-inherent contributions to a phenotype. Our findings underscore the need to analyze confounders in AI-based oncology biomarkers. To enable this, we developed a validated model applicable to other cancers and larger, diverse datasets. The German Federal Ministry of Health, the Max-Eder-Programme of German Cancer Aid, the German Federal Ministry of Education and Research, the German Academic Exchange Service, and the EU.

MRRM: Advanced biomarker alignment in multi-staining pathology images via multi-scale ring rotation-invariant matching.

Pathology image matching is crucial for assisting pathologists in the comprehensive diagnosis of cancerous areas. However, variations in image rotation and staining caused by inherent slide imaging techniques increase the burden on pathologists, complicating the examination of cancer across different pathology slides. To address this challenge, we introduce multi-scale ring rotation-in variant matching (MRRM), which improves image matching efficiency using ring topology, assisting pathologists in robustly aligning biomarker information across various pathology images. Specifically, by employing multi scale rings as convolution kernels, we accurately locate key points from the differencing of the ring pyramid, which not only enhances the likelihood of successful pathology image matching but also supports our feature descriptor in achieving advantageous performance in rotation in variance. Experiments show that with manually annotated golden landmarks as the standard in 81 cases, exhibiting significantly superior matching accuracy (130.93 μm) and a success rate of 93.83% compared to other methods, particularly in cases with rotated pathology images. This meets the routine diagnostic requirements of pathologists for cancer diagnosis.

Serum Lactate Dehydrogenase Is a Novel Predictor for the Severity in the Patients With MAFLD: A Cross-Sectional Study in Hefei, China.

The aim of this study was to assess the association between the level of lactate dehydrogenase (LDH) and the severity of metabolic syndrome (MetS) and Metabolic Associated Fatty Liver Disease (MAFLD) and the potential diagnostic value of LDH for identifying at-risk metabolic associated steatohepatitis (MASH). This cross-sectional, real-world retrospective study enrolled a total of 1118 obese patients in the department of bariatric surgery at the Second Affiliated Hospital of Anhui Medical University from January 1, 2018, to December 31, 2021. Of these, 855 were enrolled in the study cohort. MAFLD was defined as the presence or absence of fatty liver disease as suggested by histologic biopsy of liver or postoperative pathology slides, or even hematology, and meets one of the following three conditions: overweight or obesity, T2DM, and metabolic dysfunction (MetS). Serum LDH activity levels were measured in 885 patients, and logistic regression was used to analyze the relationship between LDH and metabolic syndrome and the severity of MAFLD. In the study cohort of 855 obese patients, 604 (70.6%) had MetS. Patients with MetS (214.1[209.0-219.2]) had significantly higher serum LDH levels than those without MetS (188.7[181.6-195.9]). Particularly, serum LDH level was significantly higher in subjects with hypertension, central obesity, diabetes mellitus or hyperglycemia, elevated levels of triglycerides, or reduced levels of high-density lipoprotein than in those without. Moreover, LDH concentrations were grouped according to the total number of MetS components present in each patient, with Serum LDH levels gradually increase with the total number of MetS components. The MASH subjects had significantly higher LDH levels than the other three less severe non-MASH cohorts, including normal liver, simple fatty steatosis, and B.MASH. Logistic regression showed that LDH was significantly and positively correlated with MAFLD, B.MASH, MASH, at-risk MASH, fibrosis grade ≥1, fibrosis grade ≥2 and fibrosis grade ≥3. Increased LDH levels were significantly and independently associated with the presence and severity of metabolic syndrome and MAFLD, indicating that LDH could be used as a novel biomarker and clinical predictor of severity of metabolic syndrome and MAFLD.

The effects of alpha-pinene against paracetamol-induced liver damage in male rats.

This study aims to evaluate the hepatoprotective effect of alpha-pinene against N-acetyl-p-aminophenol, paracetamol, (APA)-induced liver damage in rats. Thirty Wistar rats were divided into five groups (n = 6): Group 1: Normal (control). Group 2: APA 640 mg/kg. Group 3: alpha-pinene 50 mg/kg (APA+ αPi 50 mg/kg). Group 4: alpha-pinene 100 mg/kg (APA+ αPi 100 mg/kg). Group 5: silymarin 50 mg/kg (APA+ SIL). Alpha-pinene or silymarin was orally administered after APA administration for 14 consecutive days. This study investigated liver damage by preparing pathology slides from liver tissue. Levels of AST, ALT, ALP, total bilirubin, total antioxidant capacity (TAC), and total oxidant status (TOS) were measured. Inflammatory factors, including NF-kB gene expression and levels of IL-6 and TNF-a, were also measured. Administering alpha-pinene with APA can prevent liver damage induced by APA. Alpha-pinene can enhance TAC while reducing TOS, ALT, AST, ALP, and total bilirubin. Moreover, the results have also revealed that alpha-pinene decreases NF-kB expression, which leads to a reduction in IL-6 and TNF-a levels. It appears that alpha-pinene induces liver protective effects against APA damage by reducing the activity of liver enzymes, improving antioxidant/oxidative status, and reducing inflammation through the regulation of NF-kB and pro-inflammatory cytokines.

Application of Intravoxel Incoherent Motion in the Prediction of Intra-Tumoral Tertiary Lymphoid Structures in Hepatocellular Carcinoma.

To explore the value of intravoxel incoherent motion (IVIM) sequences in predicting intra-tumoral tertiary lymphoid structures (TLSs). This prospective study pre-operatively enrolled hepatocellular carcinoma (HCC) patients who underwent magnetic resonance imaging including IVIM sequences, between January 2019 and April 2021. Intra-tumoral TLSs presence was assessed on pathological slide images. Clinical and radiological characteristics were collected. IVIM quantitative parameters and radiomics features were obtained based on the whole delineated tumor volume. By using feature selection techniques, 22 radiomics features, clinical-radiological features (lymphocyte count and satellite nodules), and IVIM parameters (apparent diffusion coefficient (ADC_90Percentile), perfusion fraction (f_Maximum)) were selected. The logistic regression algorithm was used to construct the prediction model based on the combination of these features. The diagnostic performance was assessed using the area under the receiver operating characteristic (AUC). The recurrence-free survival (RFS) was evaluated with Kaplan-Meier curves. A total of 168 patients were divided into training (n=128) and testing (n=40) cohorts (mean age: 56.83±14.43 years; 149 [88.69%] males; 130 TLSs+). In testing cohort, the model combining multimodal features demonstrated a good performance (AUC: 0.86) and significantly outperformed models based on single-modality features. The model based on radiomics features (AUC: 0.80) had better performance than other features, including IVIM parameter maps (ADC_90Percentile and f_Maximum, AUC: 0.72) and clinical-radiological characteristics (satellite nodules and lymphocyte counts, AUC: 0.59). TLSs+ patients had higher RFS than TSLs- patients (all p <0.05). The nomogram based on the proposed model can be used as a pre-operative predictive biomarker of TLSs. The nomogram incorporating IVIM sequences may serve as a pre-operative predictive biomarker of intra-tumoral tertiary lymphoid structure (TLS) status.

Histopathology and proteomics are synergistic for high-grade serous ovarian cancer platinum response prediction

Patients with High-Grade Serous Ovarian Cancer (HGSOC) exhibit varied responses to treatment, with 20–30% showing de novo resistance to platinum-based chemotherapy. While hematoxylin-eosin (H&E)-stained pathological slides are used for routine diagnosis of cancer type, they may also contain diagnostically useful information about treatment response. Our study demonstrates that combining H&E-stained whole slide images (WSIs) with proteomic signatures using a multimodal deep learning framework significantly improves the prediction of platinum response in both discovery and validation cohorts. This method outperforms the Homologous Recombination Deficiency (HRD) score in predicting platinum response and overall patient survival. Our study suggests that histology and proteomics contain complementary information about biological processes determining response to first line platinum treatment in HGSOC. This integrative approach has the potential to improve personalized treatment and provide insights into the therapeutic vulnerabilities of HGSOC.

Prevalence of occult endometrial carcinoma in patients with endometrial intraepithelial neoplasia who underwent hysterectomy

ObjectivesTo determine the prevalence of occult endometrial carcinoma in patients with endometrial intraepithelial neoplasia (EIN) post-hysterectomy and identify pre-hysterectomy risk factors predictive of occult carcinoma.MethodsThis retrospective study included patients diagnosed with EIN between 2007 and 2021 who underwent hysterectomy as primary treatment. An expert gynecologic pathologist reviewed pathological slides. Data collected from medical records included demographic and gynecologic information, sonographic findings, and surgical and pathological outcomes. The prevalence of occult endometrial carcinoma was calculated. Descriptive statistics evaluated carcinoma incidence, and logistic regression analysis identified independent risk factors.ResultsA total of 113 patients were evaluated. The median time to hysterectomy was 9.1 weeks (range 5.8–12.8 weeks). Post-hysterectomy, 36 patients (31.8%) were diagnosed with endometrial carcinoma, all endometrioid type. Of these, 88.9% were stage I per the International Federation of Gynecology and Obstetrics classification system, and 11.1% were at high risk for nodal metastasis. Predictive factors for occult carcinoma included the intraoperative gross lesion size (2 cm or larger and less than 2 cm) and endometrial aspiration. Adjusted odds ratios were 6.723 (95% CI 2.338 to 19.333) for lesions 2 cm or larger, 3.381 (95% CI 1.128 to 10.132) for lesions less than 2 cm, and 2.752 (95% CI 1.092 to 6.936) for endometrial aspiration.ConclusionsOccult endometrial carcinoma was identified in 31.8% of patients with a pre-hysterectomy EIN diagnosis. The significant predictors were endometrial aspiration and the presence of a gross lesion during surgery.

P0750 Association between intestinal fibrosis and clinical and surgical recurrence after intestinal resection in stenotic Crohn's disease patients

Abstract Background Despite recent advances in the medical care of Crohn's disease (CD), most patients, particularly those with stenosis, still require intestinal resection during their disease course. However, more than half of the patients experience postoperative recurrence. Intestinal fibrosis is an important pathological feature of stenotic Crohn's disease. We aim to investigate the association between fibrosis at the site of intestinal stenosis and normal area and risk of clinical and surgical recurrence after first intestinal resection in patients with stenotic Crohn's disease. Methods A retrospective, multi-center study was conducted on stenotic Crohn's disease patients who received first intestinal resection between January 2010 and July 2023. The most representative pathological slides from the resected stenotic intestinal tissues and normal slides from the normal tissues were selected. We used Masson's trichrome staining to visualize collagen fibers and calculated the proportion of collagen fiber area in the slides using the Image-Pro-Plus software to assess the degree of intestinal fibrosis. Baseline information and postoperative clinical recurrence were collected through electronic medical record systems and telephone follow-up. Survival analysis and univariate and multivariate Cox regression models were used to assess the impact of intestinal fibrosis on clinical and surgical recurrence. Results 257 patients were included in this study. 94 (36.5%) and 25 (9.7%) of the patients experienced clinical and surgical recurrence during the follow-up period, respectively. In the univariate cox analysis, high degree transmural intestinal fibrosis at stenotic area was significantly associated with clinical recurrence (HR 4.94, 95%CI 2.72-8.96, P&amp;lt;0.001) and surgical recurrence (HR 4.45, 95%CI 1.45-13.72, P=0.009). This association remained significant when considering fibrosis only in the mucosa, submucosa, muscular layers of stenotic area, or in the normal area. In the multivariate cox analysis, transmural fibrosis was still related to clinical recurrence. Conclusion The degree of transmural fibrosis at the site of intestinal stenosis and normal area was significantly associated with clinical recurrence after first intestinal resection in patients with stenotic Crohn's disease. Intestinal fibrosis should be considered in postoperative treatment and follow-up decision-making.

DOP080 Artificial Intelligence-Based Prediction of Nancy Grade Activity Using Digital Pathology in Ulcerative Colitis Patients

Abstract Background Therapeutic goals of Ulcerative colitis have shifted from symptom control to achieving endoscopic and histologic remission as a key predictor of long-term outcomes. The Nancy Histological Index is widely used to assess histologic activity in UC, but it relies on subjective pathologist evaluation, leading to potential variability and increased workload. While Artificial Intelligence (AI) models have been explored for histologic assessment, few utilize comprehensive, cell-level annotations across biopsy slides. This study aims to develop AI model for predicting Nancy grades directly from whole-slide images, supporting the need for more objective, scalable, and efficient histologic grading in patients of UC. Methods A dataset of 174 digital pathology slides from UC patients, collected between January 2018 and December 2019 at Samsung Medical Center, was analyzed. A two-stage AI model was developed to predict Nancy grade. In the first stage, a U-Net-based segmentation model classified neutrophils, plasma cells, lymphocytes, and eosinophils, assessing correlation between Nancy grade and each cell type. In the second stage, predicted neutrophil counts were used to predict Nancy grade activity, comparing logistic regression (LR), random forest (RF), and extreme gradient boosting models (XGB). Model performance was evaluated using area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, and specificity. Results The segmentation model at a 200 threshold achieved a Mean IoU 0.67, and Dice score 0.70 for testing slides. Spearman correlations for activity on the Nancy grade were 0.71 for neutrophils, 0.56 for eosinophils, 0.34 for lymphocytes, and 0.31 for plasma cells, all statistically significant (P &amp;lt;0.05). The activity prediction models (neutrophils alone vs. incorporating eosinophils) achieved test the AUROCs of 0.90 vs. 0.95 for LR, 0.85 vs. 0.94 for RF, and 0.86 vs. 0.92 for XGB, respectively. The best-performing model was LR incorporating eosinophils, with an accuracy of 0.92, sensitivity of 0.88, Positive Predictive Value of 0.96, Positive Likelihood Raio of 21, and F1 score of 0.91 (Figure 1). Our model identified additional active disease in 22 out of 76 cases that had been classified as inactive by pathologists. Conclusion This study demonstrates that an AI model predicting Nancy grade can serve as a clinically valuable tool in UC management. The model, which incorporates neutrophil and eosinophil counts, achieved high accuracy and sensitivity, providing an efficient and objective method for histologic assessment. Notably, the model identified additional active cases in slides initially rated as inactive by pathologists, underscoring its potential to enhance detection accuracy in clinical settings.

Integration of clinical, pathological, radiological, and transcriptomic data improves prediction for first-line immunotherapy outcome in metastatic non-small cell lung cancer

Immunotherapy is improving the survival of patients with metastatic non-small cell lung cancer (NSCLC), yet reliable biomarkers are needed to identify responders prospectively and optimize patient care. In this study, we explore the benefits of multimodal approaches to predict immunotherapy outcome using multiple machine learning algorithms and integration strategies. We analyze baseline multimodal data from a cohort of 317 metastatic NSCLC patients treated with first-line immunotherapy, including positron emission tomography images, digitized pathological slides, bulk transcriptomic profiles, and clinical information. Testing multiple integration strategies, most of them yield multimodal models surpassing both the best unimodal models and established univariate biomarkers, such as PD-L1 expression. Additionally, several multimodal combinations demonstrate improved patient risk stratification compared to models built with routine clinical features only. Our study thus provides evidence of the superiority of multimodal over unimodal approaches, advocating for the collection of large multimodal NSCLC datasets to develop and validate robust and powerful immunotherapy biomarkers.

Label credibility correction based on cell morphological differences for cervical cells classification

Cervical cancer is one of the deadliest cancers that pose a significant threat to women’s health. Early detection and treatment are commonly used methods to prevent cervical cancer. The use of pathological image analysis techniques for the automatic interpretation of cervical cells in pathological slides is a prominent area of research in the field of digital medicine. According to The Bethesda System, cervical cytology necessitates further classification of precancerous lesions based on positive interpretations. However, clinical definitions among different categories of lesion are complex and often characterized by fuzzy boundaries. In addition, pathologists can deduce different criteria for judgment based on The Bethesda System, leading to potential confusion during data labeling. Noisy labels due to this reason are a great challenge for supervised learning. To address the problem caused by noisy labels, we propose a method based on label credibility correction for cervical cell images classification network. Firstly, a contrastive learning network is used to extract discriminative features from cell images to obtain more similar intra-class sample features. Subsequently, these features are fed into an unsupervised method for clustering, resulting in unsupervised class labels. Then unsupervised labels are corresponded to the true labels to separate confusable and typical samples. Through a similarity comparison between the cluster samples and the statistical feature centers of each class, the label credibility analysis is carried out to group labels. Finally, a cervical cell images multi-class network is trained using synergistic grouping method. In order to enhance the stability of the classification model, momentum is incorporated into the synergistic grouping loss. Experimental validation is conducted on a dataset comprising approximately 60,000 cells from multiple hospitals, showcasing the effectiveness of our proposed approach. The method achieves 2-class task accuracy of 0.9241 and 5-class task accuracy of 0.8598. Our proposed method achieves better performance than existing classification networks on cervical cancer.

Galectin-3 marker expression in renal cell carcinoma and correlation with patient's clinicopathologic factors: A cross-sectional study.

Renal Cell Carcinoma is one of the most common malignancies worldwide. To date, multiple attempts had been made to accurately diagnose it and predict its behavior. One of the most intriguing biomarkers that have been assessed since 90s, is galectin-3. This study aimed to increase the prior knowledge of galectin-3 expression association with patient's clinicopathologic factors. In this single-center cross-sectional study, 71 patient samples from hospital archive were assessed for galectin-3 expression by immunohistochemistry assay. Pathologic slides were evaluated for histologic subtype, grade, stage, tumor size, presence of necrosis, and invasion of the renal vein. By adding cytoplasmic staining score to the color intensity score, a final score was recorded as galectin-3 positivity score. 88 pathological slides of patients with confirmed RCC were screened and 71 were finally assessed. The mean age of the patients was 58.52 years (lowest 30 and highest 87). 67.6% were males and 32.4% were females. 68% of tumors were clear cell carcinoma, and only one oncocytoma was present. All 9 chromophobe cases showed a strong galectin-3 expression. Except for female gender (47.8% vs 18.8% in men; P=0.01), no statistically significant association was found between patient age, tumor grade, tumor size, tumor stage, and renal vein invasion with the level of galectin-3 expression. Considering the contradictory findings between this study and other similar studies, it can be concluded that the physiological role of galectins is very complex and the need for larger and more comprehensive evaluations is felt.