Research square最新文献

{"title":"Clinically Contextualized State Spaces Provide Continuous Representations of Sepsis.","authors":"Seibi Kobara, Morgan Kay, Tilendra Choudhary, Vijay Krishnamoorthy, Mihai V Podgoreanu, Yoshiro Hayashi, Suresh M Agarwal, Rishikesan Kamaleswaran","doi":"10.21203/rs.3.rs-11221040/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-11221040/v1","url":null,"abstract":"<p><p>Precision medicine increasingly leverages molecular and clinical similarity to identify patient subgroups and guide individualized care. However, conventional cross-sectional approaches often characterize biological states independently of the clinical context in which they arise. Similar laboratory profiles may have different clinical implications depending on subsequent trajectory, while biologically distinct states may share similar clinical outcomes. Consequently, clustering based solely on clinical trajectory-agnostic clustering may incompletely capture clinically relevant heterogeneity. To address these limitations, we introduce CERL (Contextual Endotype and State Representation Learning), a clinical trajectory-informed representation learning framework that integrates clinical contexts to construct patient-state representations from clinical data. Rather than defining patients solely through discrete clusters, CERL learns a continuous clinical state space that preserves biological heterogeneity while organizing states according to clinically relevant trajectories. Across both simulated and real datasets, CERL robustly identified latent patient groups and reliably reproduced the learned patient groups in validation. In a proof-of-concept application across multi-institutional critical care cohorts, CERL identified three reproducible clinical state groups embedded within a state representation. The representation captured systematic variation in routine laboratory phenotypes and demonstrated clinical trajectory-specific redistribution dynamics within the continuous state space. These findings establish CERL as a framework for generating clinically contextualized state representations that integrate biological characteristics with clinical trajectories, providing a foundation for longitudinal pathological progression modeling and future digital twin approaches in critical care.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692580/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280077","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Neuromedin U receptor 2 agonists suppress cocaine self-administration in male rats.","authors":"Vena Martinez, Hong Sun, Cun Zhang, Robert Fox, Jia Zhou, Kathryn Cunningham, Jonathan Hommel","doi":"10.21203/rs.3.rs-11000092/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-11000092/v1","url":null,"abstract":"<p><p>Rationale Cocaine use disorder ( <b>CUD</b> ) remains without an FDA-approved pharmacotherapy, highlighting the need for novel mechanistically distinct targets. Prior work implicates neuromedin U ( <b>NMU</b> ) and its receptor NMU receptor 2 ( <b>NMUR2</b> ) in cocaine-directed behavior, but NMUR2 peptide-based approaches have limited translational utility. Objectives Here, we tested whether the previously characterized NMUR2-active small molecules NY0116 and NY0128 reduce cocaine self-administration under a progressive ratio schedule in male rats. Because reduced progressive ratio responding can also reflect broader behavioral disruption, we examined response latency, inactive lever responding, and locomotor activity. Methods Rats were trained to self-administer intravenous cocaine (0.25 mg/kg/infusion) and tested after systemic administration of NY0116 or NY0128. Results Both compounds reduced cocaine-maintained responding at the highest dose tested, including reductions in breakpoint and active lever responding. Lever discrimination was not significantly altered at the effective dose, arguing against loss of task discrimination as the explanation for reduced cocaine intake. Individual-animal analyses showed no significant correlation between levels of locomotor activity and cocaine infusions during NY0128 test sessions, indicating that locomotor output did not account for individual differences in cocaine intake. Conclusions These findings demonstrate that NY0116 and NY0128 are effective in reducing cocaine self-administration in male rats, providing strong evidence that NMUR2 is a promising target for developing new CUD pharmacotherapies.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692601/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280146","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Single-Cell Multiplex Immune Biomarker Prediction from Histopathology Across Cancer Types and Clinical Endpoints.","authors":"Usman Afzaal, Usama Sajjad, Abdul Akbar, Tianyang Wang, Elshad Hasanov, Sachin Jhawar, Wei Chen, Zaibo Li, Muhammad Khalid Khan Niazi","doi":"10.21203/rs.3.rs-11002687/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-11002687/v1","url":null,"abstract":"<p><p>Background The spatial organization of cellular lineages within the tumor microenvironment governs disease progression and therapeutic response, but multiplexed spatial profiling remains costly, technically complex, and tissue-destructive. Routine hematoxylin and eosin (H&E) histopathology is inexpensive and widely available but lacks explicit molecular information. This study aims to determine whether single-cell, biomarker-informed cellular states can be inferred directly from H&E images to enable scalable, non-destructive spatial biology across cancer types and clinical endpoints. Methods We present CellPlexer, a deep learning framework trained on more than 36 million co-registered cells from paired H&E and multiplex immunofluorescence images to jointly segment nuclei and classify each cell into one of 15 hierarchically defined biomarker states. A spatially aware transformer integrates single-cell embeddings with patch-level histomorphology through attention-based multimodal fusion, generating a multi-scale whole-slide representation for clinical prediction. We validated CellPlexer across four independent cohorts spanning lung, breast, and colorectal cancer. Results Here we show that CellPlexer consistently outperformed morphology-only baselines across recurrence, treatment response, survival, and mutation prediction tasks, achieving an AUC of 72.19% for pathological complete response prediction in triple-negative breast cancer (75.06% with clinical variable integration) and 83.09% for BRAF mutation prediction in colorectal cancer. Conclusions These results demonstrate that morphologically inferred, single-cell biomarker states capture clinically and biologically meaningful tumor microenvironment organization, providing a scalable, non-destructive framework for extracting proteomic-level insights from widely available clinical histopathology archives.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692624/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280297","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Quantitative gene-level inference of tumor transcription from plasma epigenomes.","authors":"Matthew Freedman, Gunsagar Gulati, Damien Vasseur, Rashad Nawfal, Shahabeddin Sotudian, Karl Semaan, Marc Eid, Ji-Heui Seo, Shin Saito, Noa Phillips, John Canniff, Hunter Savignano, Surya Chhetri, Zhenjie Jin, Yukun Ou, Mohamed-Amine Bani, Gwo-Shu Lee, Rachel Trowbridge, Ilana Epstein, Gabriella Rickards, Paulo Cordeiro, Razane El Hajj Chehade, Renchin Wu, Maria-Anna Chrysovergi, Ze Zhang, Brady James, Massard Christophe, Antoine Italiano, Antoine Hollebecque, Jean-Charles Soria, Fabrice André, Cécile Badoual, Joaquim Bellmunt, Harshabad Singh, Andrew Aguirre, Brian Wolpin, Jacob Sands, Matthew Oser, Toni Choueiri, Sylvan Baca","doi":"10.21203/rs.3.rs-11043657/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-11043657/v1","url":null,"abstract":"<p><p>Tumor phenotype and therapeutic vulnerability are shaped by dynamic transcriptional states that are not captured by genetic profiling alone and are difficult to monitor longitudinally. Existing plasma-based approaches have largely been developed using samples with high circulating tumor DNA fractions, evaluated across expression quantiles rather than individual genes, and rarely benchmarked head-to-head. Using timepoint-matched plasma epigenomes and tumor transcriptomes, we identify biologically grounded relationships between circulating chromatin, fragmentation, and transcription. H3K4me3 and H3K36me3 encode distinct positional information reflecting promoter activity and transcriptional elongation, while gene-body fragment periodicity predicts expression. We integrate these relationships into APEX, a machine-learning framework for genome-wide single-gene expression inference. Across 559 plasma samples spanning 24 cancer types and clinically relevant tumor fractions, APEX outperforms existing approaches for individual-gene inference, resolves pancreatic cancer transcriptional subtypes, tracks therapeutic targets in small-cell lung cancer, and identifies NECTIN4 expression as a biomarker of response to enfortumab vedotin in urothelial cancer.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692620/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280185","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Increase in Methadone Treatment Episodes With Co-occurring Mental Health Disorders between 2013 and 2023.","authors":"Yinfei Kong, Kechen Zhao, Alanis Goncalves Leite, Daniel Howard","doi":"10.21203/rs.3.rs-10654615/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-10654615/v1","url":null,"abstract":"<p><p>Objectives To compare US methadone treatment episodes with co-occurring mental health disorders in 2013 and 2023. Methods We analyzed 2013 and 2023 Treatment Episode Data Set-Discharges records for 75,338 episodes with methadone use, a co-occurring mental health disorder, valid sex or gender information, and discharge because treatment was completed, dropped out, or terminated by the facility. We compared demographics, substance use, treatment, and discharge characteristics. Results Eligible episodes increased from 18,935 in 2013 to 56,403 in 2023. Compared with 2013, 2023 episodes involved older clients (mean age, 39.72 vs 36.16 years), fewer female clients (40.2% vs 49.8%), more methamphetamine/speed use (14.7% vs 3.2%), more cocaine/crack use (38.8% vs 29.9%), less injection drug use (37.1% vs 55.3%), more Medicaid payment (22.9% vs 7.7%), and more homelessness (19.3% vs 12.7%). Conclusions The profile of methadone treatment episodes with psychiatric comorbidity changed markedly. Public Health Implications: Treatment systems may need greater capacity for stimulant co-use, aging-related needs, Medicaid-financed integrated care, and housing support.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692592/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280200","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Modifiable health-factor burden and stroke outcomes across three population databases: cross-sectional and longitudinal evidence from NHANES, CHARLS and HRS.","authors":"Qingliang Han, Wenliu Huang, Xianqiong Tan, Lingdang Zhang, Qinglin Feng, Wenwu He","doi":"10.21203/rs.3.rs-10727783/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-10727783/v1","url":null,"abstract":"<p><p>Background Modifiable vascular and lifestyle factors often cluster in middle-aged and older adults. We examined whether a simple harmonized health profile provides comparable stroke-related information across public population databases. Methods We used NHANES 2007-2018, CHARLS 2018 and HRS 2018 for cross-sectional analyses, and CHARLS 2011-2018 and HRS 2014-2022 for longitudinal validation. The cross-sectional outcome was self-reported physician-diagnosed stroke history. The main exposure was a six-factor burden score covering physical inactivity, hypertension, diabetes, dyslipidemia or high cholesterol, heart disease and current smoking. Database-specific weighted logistic regression estimates were pooled using random-effects meta-analysis. Longitudinal analyses used person-wave discrete-time survival models. Prediction analyses used CHARLS for model development and HRS for external validation. Results The adjusted cross-sectional models included 14,475 NHANES participants, 16,070 CHARLS participants and 8,175 HRS participants. Stroke history increased monotonically across burden categories in all three databases. Each additional unfavorable factor was associated with higher odds of stroke history in NHANES (odds ratio, 1.69; 95% confidence interval, 1.55-1.86), CHARLS (1.67; 1.58-1.77) and HRS (1.63; 1.52-1.76). The pooled odds ratio was 1.66 (95% confidence interval, 1.60-1.73), with little observed heterogeneity. In longitudinal models, each 1-point higher brain-health score was associated with lower hazards of incident self-reported stroke in CHARLS (hazard ratio, 0.740; 95% confidence interval, 0.712-0.769) and HRS (0.830; 0.789-0.874). The brain-health score showed moderate discrimination in CHARLS (area under the curve, 0.645) and HRS (0.648). Conclusions A higher burden of unfavorable modifiable health factors was consistently associated with stroke history across three public datasets. Longitudinal analyses linked a higher brain-health score to lower incident self-reported stroke hazards. These profiles may support low-cost population risk stratification, but they do not support causal or individual-level clinical interpretation.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692589/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280171","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TransPrimer: A Web-Based Tool for Automated Transcript-Specific RT-PCR/qPCR Primer Design Across Multiple Species.","authors":"Junyeop Ahn, Yun H Lee, Donald J Zack, Ping-Wu Zhang","doi":"10.21203/rs.3.rs-11225556/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-11225556/v1","url":null,"abstract":"<p><p>Transcript-specific RT-PCR/qPCR primer design is challenging when closely related RNA variants share extensive exonic sequence and differ primarily in exon connectivity. We developed TransPrimer, a web application for automated transcript-specific and pan-transcript primer design. TransPrimer retrieves curated RefSeq protein-coding and non-coding transcripts for six species and compares sibling transcripts to identify discriminating exons and exon-exon junctions (EEJs). The application automatically determines whether transcript specificity can be achieved using exon-targeted primers or requires single- or dual-EEJ strategies. A complementary pan-transcript mode identifies primer pairs that amplify all annotated transcripts of a gene, when feasible, or the largest compatible subset. Junction-spanning primers are evaluated using an arm-specific thermodynamic framework that independently calculates melting temperatures (Tm) for the full primer and its two exon-matching arms using nearest-neighbor thermodynamics and mixed-salt correction. Primer3-based criteria are integrated for partner-primer design and evaluation. User-adjustable reaction parameters, interactive transcript visualization, and custom sequence input support flexible assay development, including novel or externally annotated transcripts. By integrating transcript retrieval, structural comparison, strategy selection, thermodynamic evaluation, and primer generation, TransPrimer reduces manual transcript inspection and facilitates transcript-specific and gene-level RT-PCR/qPCR assay development across diverse transcript architectures.URL: https://transprimer.vercel.app/.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692623/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280266","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Maternal transmission and limited virulent phage replication shape early infant gut virome maturation.","authors":"Tommi Vatanen, Kim Kreuze, Lilia Björkgren, Christopher Turkington, Neda Nezam-Abadi, Sam Seymour, Andrey Shkoporov, Colin Hill, Ville Friman, Mikael Knip","doi":"10.21203/rs.3.rs-11232673/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-11232673/v1","url":null,"abstract":"<p><p>Early life gut microbiome dynamics are increasingly recognized as important determinants of microbiome maturation, yet the ecological dynamics of its associated bacteriophages remain poorly resolved. Here, we performed bulk and virus like particle (VLP)-enriched metagenomic sequencing and fecal metabolomic profiling of longitudinal fecal samples from 32 mother-infant dyads, incorporating an exogenous phage spike in to enable absolute quantification of viral loads. Our results show that early infant gut phageome is defined by a profound lack of virulent phage genome copies while temperate phages remain relatively stable, challenging previous assumptions. We further identified extensive maternal sharedness of mainly virulent phages predicted to infect Bacteroidaceae, which could explain recovery of virulent phage absolute abundances. Virulent phage concentrations were further associated with several fecal metabolites including primary bile acids. These findings revise the current model of infant phageome assembly highlighting the underappreciated role of virulent phages.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692616/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280204","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Cross-Band Coupling of EEG Infraslow Oscillations during NREM Sleep in Mice.","authors":"Max Tan, Oscar Ramirez-Plascencia, Wei Wang, Amanda Studnicki, J Kent Werner, Elizaveta Tremsina, Shiqian Shen, Can Zhang, Tingying Helen Zeng, Robert Thomas, Haoqi Sun","doi":"10.21203/rs.3.rs-10985934/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-10985934/v1","url":null,"abstract":"<p><p>The infraslow oscillation (ISO) is a ~ 0.02 Hz rhythmic fluctuation in electroencephalographic (EEG) power during NREM sleep. Prior studies were limited to sigma-band ISO, linked to rhythmic norepinephrine release in mice. Whether ISOs exist in other bands, and how ISOs across bands are coupled, remain unknown. We analyzed cross-band ISO coupling across six frequency bands during NREM using a wild-type (WT) discovery dataset of 122 recordings from 72 mice and WT validation of 7 mice. A transgenic (TG) validation of 6 mice was added to test ISO coupling's robustness to Cre knock-in on the LepR gene (LepR-Cre). Band power time series were filtered to the ISO range of 0.01-0.03 Hz. Cross-band coupling was assessed using Spearman's correlation and phase-locking value (PLV) across all 15 band pairs. ISOs were strongest in alpha, sigma, beta, and theta bands, weak in gamma, and absent in delta. Alpha, sigma, beta, and theta ISOs formed a highly correlated motif with replication in WT and TG (r = 0.61-0.77), while gamma had moderate correlation (r = 0.30-0.42), and delta was uncorrelated (p > 0.05). All band pairs showed significant phase locking with replication (FDR-corrected q < 0.01), with stronger consistency in alpha, sigma, theta, and beta (PLV 0.54-0.71), and weaker consistency involving delta and gamma (PLV 0.12-0.37). EEG ISOs extend beyond the sigma band to alpha, beta, and theta bands, forming a strongly coupled motif that may reflect a fundamental organization principle of sleep microstructures at a minute scale.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692581/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280096","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Detection of extraprostatic extension of prostate cancer by transfer learning AI models versus radiologists on biparametric and multiparametric MRI.","authors":"Angela Tong, Alexander Daniels, Luke Ginocchio, Paul Smereka, Tarun Dutt, Lavanya Umapathy, Madhur Nayan, Hersh Chandarana, Patricia Johnson","doi":"10.21203/rs.3.rs-11001056/v1","DOIUrl":"https://doi.org/10.21203/rs.3.rs-11001056/v1","url":null,"abstract":"<p><p><b>Objectives</b> To develop and test deep-learning models which detect extraprostatic extension (EPE) on biparametric (bpMRI) and multiparametric MRI (mpMRI) of the prostate in comparison to radiologists. <b>Methods</b> Consecutive patients at a large healthcare enterprise who underwent prostate MRI (2015 to 2023) with subsequent radical prostatectomy within 1 year were included. The dataset was divided into training/validation/test sets. Transfer learning models, composed of two multi-branch 3D convolutional neural networks based on a 3D ResNet-50 backbone, were trained on bpMRI (AI _bp ) with an input of axial T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI) and on mpMRI (AI _mp ) with an input of axial T2WI, axial DWI, and axial post-contrast imaging. No prostate gland or tumor segmentations were performed. A logistic regression model with prostate specific antigen density was also trained with the AI _mp model (AI _mpPSAD ). Three fellowship-trained abdominal radiologists evaluated the test set for EPE on a 1 to 5 scoring system based on capsular appearance, on bpMRI and mpMRI for each study. Areas under the receiver operating characteristic curves (AUC) were obtained and compared with Delong test. Diagnostic statistics were also obtained. <b>Results</b> 1,232 prostate MRIs were included (1003/113/116 in the training/validation/test sets). AUC for models were as follows: AI _bp (0.71), AI _mp (0.72), and AI _mpPSAD (0.73). While holding sensitivity constant at the Youden index for AI _bp (0.77), specificity and accuracy of the AI _bp , AI _mp , and AI _mpPSAD were 0.64, 0.69, 0.71 and 0.71, 0.73, 0.74, respectively. The ranges of AUC for readers were: bpMRI (0.66-0.71) and mpMRI (0.69-0.74). The ranges of accuracy of the readers on bpMRI was 0.60-0.68 and on mpMRI was 0.65-0.72. <b>Conclusion</b> A segmentation-free transfer learning model performs comparably to fellowship-trained radiologists on the detection of EPE on both bpMRI and mpMRI. There was a trend of increasing accuracy with the addition of post contrast sequences for all models and readers.</p>","PeriodicalId":519972,"journal":{"name":"Research square","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13692593/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149280039","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}