Microsatellite instability (MSI) is an important prognostic and predictive biomarker in colorectal carcinoma and is associated with several characteristic histomorphological features. While large language models (LLMs) have been investigated for structured data extraction from histopathology reports, their ability to infer molecular genotype from text-based morphologic descriptions remains underexplored. This study evaluated structured data extraction, MSI prediction performance, confidence behavior, and biological reasoning quality of two LLMs using colorectal carcinoma pathology reports with The Cancer Genome Atlas (TCGA)-assigned MSI status as the reference standard. Fifty colorectal carcinoma reports, including 25 MSI-high (MSI-H) and 25 microsatellite-stable (MSS), from TCGA database were stratified by morphologic signal strength. ChatGPT 5.2 (OpenAI, San Francisco, CA) and Gemini 3 (Google, Mountain View, CA) were given an identical zero-shot prompt to extract 17 histopathologic variables across structural, staging, invasion, and MSI-associated morphologic categories, predict MSI status with confidence scores, and provide biological justification. Both models demonstrated high extraction accuracy for structural variables (85.0-88.5%) but substantially lower performance for MSI-associated morphologic variables (ChatGPT 5.2: 39.4%; Gemini 3: 43.9%), with mucinous differentiation showing the highest omission rate and signet ring cell morphology the lowest extraction accuracy. Overall, MSI prediction accuracy was 64.0% for Gemini 3 and 16.0% for ChatGPT 5.2. Among committed predictions, ChatGPT 5.2 demonstrated perfect specificity and positive predictive value, whereas Gemini 3 showed a morphologic signal-dependent accuracy gradient and a substantially higher rate of high-confidence incorrect predictions (20.0% versus 2.0%). General-purpose LLMs demonstrated variable ability to recognize MSI-associated histomorphologic descriptors, with systematic failures in extraction directly limiting downstream genotype inference.
 Artificial intelligence (AI) tools like ChatGPT (OpenAI, San Francisco, USA) are increasingly influencing medical education, yet their use among undergraduate students in Western India remains underexplored. This study aimed to assess students' acceptance, usage patterns, and key factors influencing AI adoption to support effective curriculum integration. This cross-sectional, questionnaire-based study evaluated attitudes toward and use of AI among undergraduate medical students from first year to final year. A structured 37-item English questionnaire was developed via Google Forms (Google LLC, Mountain View, USA) and distributed online using electronic channels, with informed consent obtained. Data were managed in Microsoft Excel 2013 (Microsoft Corp., Redmond, USA) and analyzed using GraphPad Prism version 10.6 (GraphPad Software, Boston, USA; www.graphpad.com), yielding key insights into perceptions and use of AI. This study was conducted among 791 medical students, with a mean age of 19.92 ± 1.64 years. Of these, 701 (88.62%) participants engaged with AI, relying predominantly on social media (425 (53.72%)) and faculty lectures (199 (25.15%)) for information. Moreover, 784 (99.11%) participants showed high AI awareness. Multivariable logistic regression demonstrated that AI use was strongly associated with behavior (odds ratio (OR)=356.9; 95% confidence interval (CI): 77.83-6329), perceived usefulness (OR=18.5; 95% CI: 40.68-327), and inversely with perceived risk (OR=0.06; 95% CI: 0.003-0.31). Students more than 20 years of age, male participants, and consistent AI users showed significantly higher usefulness and strong positive attitudes. Notably, first-year students reported a higher perception of risk compared to their peers, underscoring significant variations in AI adoption and perception.  The use of AI models among Western Indian undergraduate medical students and their attitudes were investigated in this study. Although there was variation in students' comprehension of AI-related concepts, the majority of students reported being aware of and having previously used AI tools, mostly through social media platforms. The use of AI and opinions of AI models were correlated with age, gender, academic year, perceived utility, and behavior. The results demonstrate the potential value of responsible and ethical AI guidance in medical education and may offer initial institutional-level insights for future multicentric research and curriculum development.
Professional society patient education materials frequently exceed recommended literacy levels, limiting equitable health information access. This study aimed to compare the readability, information quality, understandability, and actionability of artificial intelligence (AI)-generated patient education materials versus professional society materials across gastroenterology, surgery, ophthalmology, and anesthesiology. We conducted a cross-sectional comparative analysis of 100 paired topics (25 per specialty), comparing professional society materials with the responses generated by ChatGPT (OpenAI, San Francisco, California, United States) under standardized conditions. Readability was assessed using the Flesch-Kincaid grade level, information quality with DISCERN, and understandability and actionability with the Patient Education Materials Assessment Tool (PEMAT). Paired two-sided t-tests assessed within-specialty differences. In surgery, AI-generated materials had lower reading levels and higher quality, understandability, and actionability (all p<0.001). In anesthesiology, AI materials were more readable (p<0.001) with no differences in other measures. In ophthalmology, AI improved readability (p<0.001), while professional society materials had higher quality and understandability (p<0.01) with no difference in actionability. In gastroenterology, AI materials had higher reading levels (p<0.001) with no differences in quality or usability. The performance of AI-generated patient education materials varied by specialty and appeared to depend on the structure and complexity of clinical content. AI improved readability in several domains, but these gains were not uniform across specialties, particularly in areas requiring more complex or longitudinal explanations.
Chemotherapy resistance in acute myeloid leukemia (AML) remains a major clinical challenge. Integration of multiomic profiling and in vivo functional genomics revealed splicing dysregulation as a determinant of chemoresistance in AML. We uncovered a network involving the splicing regulator SRRM1 and the CLK1/4 and PAK1 kinase families as vulnerabilities in chemoresistant AML cells. Both kinase families are hyperactivated in chemoresistant cells, promoting SRRM1 phosphorylation and altering its scaffolding function. We also identified a relapse-associated PAK1 variant, c.1429G>T p.(Ala477→Ser), that confers chemotherapy resistance. Combined PAK1 and CLK1/4 inhibition recapitulated the splicing changes induced by SRRM1 loss, preferentially targeting chemoresistant AML and enhancing chemotherapy efficacy in cell lines, primary cells, and mouse models. Last, we pinpointed MAP2K5 as a critical downstream effector because missplicing of exons 17 and 18 of MAP2K5 upon SRRM1 depletion sensitized cells to chemotherapy. Our findings highlight a therapeutic strategy to overcome AML relapse by targeting splicing dysregulation.
Quadratic attention enhances interaction capacity in Transformer models but leads to rapid growth in computational demands as attention rank increases. This paper presents a bounded-rank quadratic attention mechanism where fixed-dimensional feature encodings determine the interaction space and enforce a strict upper bound on attention rank. A Fourier-domain formulation offers a spectral interpretation of the quadratic kernel via unitary transformation while maintaining exact attention computation. The proposed approach achieves bounded-rank attention with computational complexity that scales linearly with sequence length when the encoding dimension remains constant. Experimental validation on a real-world Tea pest image dataset yields 84.5% classification accuracy, surpassing the 82.1% achieved by a standard Vision Transformer under equivalent experimental conditions. Attention processing time per image declines from 14.8 ms to 5.6 ms. Peak memory usage declines from 820 MB to 430 MB. Although memory bandwidth and kernel launch overhead restrict the measured speedup to [Formula: see text], the accuracy loss under additive noise reaches 2.9% compared to 3.1% for the baseline, demonstrating comparable robustness. These findings indicate that explicit rank control in attention mechanisms can be realized through representational design, offering an efficient bounded-rank alternative to conventional full-rank attention.
Human iPSC-derived hepatocytes are widely used in disease modeling. However, late embryonic development in the human liver remains elusive, which hinders differentiation. During late liver embryonic development, Topoisomerase II (TOP2) is downregulated; however, its role in differentiation is unclear. We replicated the TOP2 silencing at birth and identified a transcription factor crucial for hepatocyte differentiation in vitro. Subtoxic inhibition of TOP2 reduces nuclear chromatin condensation without causing DNA damage. RNA-seq analysis revealed that TOP2 inhibition induced cell cycle arrest, accompanied by FOXM1 downregulation. ATAC-seq confirmed that TOP2A inhibition decreased chromatin accessibility and modulated the Wnt/β-catenin pathway. Proteomic analysis demonstrated that FOXM1 inhibition mimicked TOP2A-mediated cell cycle arrest and reduced the levels of fetal hepatocyte proteins. Prolonged FOXM1 inhibition results in increased hepatocyte polyploidization, enhanced CYP450 activity, and improved lipid metabolism. Our findings suggest that FOXM1 inhibition promotes terminal differentiation of human iPSC-derived hepatocytes, potentially paving the way for understanding late embryonic development in the human liver.
暂无摘要（点击查看详情）
Plasmodium falciparum and Plasmodium vivax cause most malaria cases worldwide and are co-endemic in many countries, yet differ substantially in biology and in their responses to common interventions. As programmes drive down P. falciparum, P. vivax is a growing challenge for elimination, but most modelling tools assess the species separately, limiting coordinated policy. We aimed to build a unified malaria transmission modelling framework for co-endemic settings and assess how well it reflects global prevalence. We integrated an established P. vivax model into a flexible P. falciparum modelling platform, enabling parallel simulation of both species within a shared biological, demographic, and intervention environment. Modelled equilibrium prevalences, matched by mosquito density, were compared with 19,225 yearly co-prevalence estimates from the Malaria Atlas Project (769 sub-national regions, 33 co-endemic countries, 2000-2024); uncertainty was represented by 95% quantile-based regions from 50 parameter draws. We assessed how this fit was modified by biological factors and simulated interventions. Here we show that the framework captures 51% of co-prevalence estimates within its uncertainty regions, rising to 65.5% when country-specific P. vivax relapse rates and human Duffy negativity are included. P. falciparum predominates where mosquito densities are high, whereas P. vivax is relatively more prevalent at lower densities. Simulated interventions produce larger relative reductions in P. falciparum prevalence, while P. vivax shows greater rebounds after intervention withdrawal, particularly at low mosquito densities. This unified framework provides a quantitative tool to support coordinated, species-specific intervention strategies in co-endemic settings, a step toward sustainable malaria elimination.
暂无摘要（点击查看详情）
Bad actors and bots present threats to data integrity in remote research designs, compromising estimates of cognitive task performance. Here, we share our experiences and recommendations for identifying bad actors and bots and preventing them from enrolling in remote studies of cognitive task performance. As part of a separate large-scale validation study, 3488 participants were recruited through various platforms, including social media and MTurk, and were compensated for their participation. When inspecting these data, we observed a high incidence of unusual or nonsensical responses to demographics questions in combination with unusual cognitive performance patterns. In response, participants were categorized according to bad actor status as determined by 1) number and risk-level associated with red flags within initial screening and pattern of responses; and 2) a researcher review process to identify patterns of aberrant, patterned, or inconsistent responding. Participants were either automatically labeled as bad actors (n = 2829), labeled as bad actors after researcher review (n = 242), or labeled as good actors after researcher review (n = 417). Cognitive task performance was assessed with Adaptive Cognitive Evaluation-Explorer (ACE-X), a mobile, gamified collection of executive function tasks including measures of working memory, attention, and cognitive flexibility. We considered differences between good and bad actors for mean response times, variability in response times, accuracy, and number of trials attempted. Across 6 of the 7 ACE-X tasks evaluated, bad actors responded slower than good actors on average (β = 34.10-150.15, p < 0.001-0.047). Additional differences in task performance suggested that overall, bad actors also tended to be more variable in their response times (β = -0.32-1.01, p < 0.001), less accurate (β = -2.15-0.00, p < 0.001- > 0.999), and responded to fewer trials (β = -1.30-0.00, p < 0.001- > 0.999). Findings presented here suggest that bad actors show consistent differences in their patterns of performance on cognitive assessments, and therefore careful consideration related to study design and protocols to ensure data integrity are needed when evaluating cognitive performance in remote research settings.
Language discordance in surgical care is a structural driver of inequity that affects patient safety, trust, and outcomes. Emerging interpreter technologies, including artificial intelligence (AI) and remote video interpretation (RVI), are rapidly entering clinical settings. However, implementation decisions are often made without understanding how patients themselves perceive these modalities or whether they view them as replacements or complementary tools within their care. To explore Spanish-speaking surgical patients' perceptions of AI- and RVI-based interpreter technologies, and to understand how clinical context influences modality preferences, the author team conducted a descriptive concurrent mixed-methods study within a U.S. academic health system, enrolling 23 adult patients with Spanish language preference across the surgical continuum. The patients did not choose a single preferred modality; instead, they expressed context-dependent needs. AI was viewed as advantageous for its speed, privacy, and literal translation in straightforward or time-sensitive scenarios. RVI was favored for emotionally complex conversations and cultural nuance. Across narratives, patient agency emerged as a dominant theme. These findings support the development of a multifaceted language access infrastructure in which AI and remote human interpreters are deployed synergistically based on clinical sensitivity, urgency, and patient preference.
Gastropleural fistulas (GPFs) are rare pathologic communications between the stomach and the pleural cavity and are associated with high morbidity and limited success with conventional surgical repair. We aimed to describe the successful management of a complex, refractory GPF using a multimodal endoscopic approach. A 27-year-old man with traumatic diaphragmatic rupture and multiple failed surgical repairs was referred to our tertiary endoscopy center. Endoscopic evaluation confirmed a gastric-to-pleural fistula. A sequential strategy was used: initial endoscopic negative pressure therapy using a modified nasogastric vacuum system, followed by closure with an over-the-scope (OTS) clip. Negative pressure therapy promoted granulation tissue formation and reduction of fistula output. Deployment of an OTS clip achieved complete sealing of the orifice, with no further thoracic drainage. Total endoscopic therapy lasted 28 days. The patient was discharged in good clinical condition. At 6-month follow-up, he remained asymptomatic, with endoscopy confirming a well-healed scar and no recurrence. Multimodal endoscopic therapy combining negative pressure therapy and OTS clip placement represents a safe and minimally invasive option for complex GPFs refractory to surgery. This case highlights the growing role of advanced endoscopic techniques in the management of challenging gastrointestinal fistulas.
Targeted dream incubation (TDI) is a highly effective method for eliciting hypnagogic dreams related to specific topics through the presentation of verbal prompts and serial awakenings at sleep onset. In this pilot study, we tested whether TDI at sleep onset can effectively direct dream content in subsequent rapid eye movement (REM) sleep. We allowed participants a daytime nap opportunity following TDI at sleep onset. Serial awakenings were performed both at sleep onset and after entry into REM sleep. Our primary objective was to assess whether the TDI protocol during the sleep onset period would continue to affect dream content in REM sleep, producing dreams of the target content ("tree") in the first REM awakening. Our second objective was to assess incorporation when participants received additional TDI prompts following REM awakenings. All 11 participants successfully incubated the target theme at sleep onset, and eight subsequently obtained REM sleep. Four of these participants (50%) incorporated the target theme into their first REM dream, and five incorporated the target theme in subsequent REM dreams (63%). Results provide preliminary evidence that TDI may impact dreams in REM sleep. This method of engineering dreams across sleep stages may be useful for understanding how dream generation and function may be continuous or different across sleep stages.
The Zika epidemic was first reported in Brazil in 2015. Since then, several factors have sustained the virus transmission chain in the country. This study aimed to describe the epidemiological profile and temporal trends, as well as to identify significant spatial and spatiotemporal clusters of Zika in Brazil between 2016 and 2023. Ecological study including confirmed Zika cases in Brazil from 2016 to 2023. Data were extracted from the National Notifiable Diseases Information System. Sociodemographic variables (age, sex, race/ethnicity, education level, and pregnancy status) were collected, along with incidence rates. Analyses were conducted in three steps: (1) description of the epidemiological profile; (2) temporal analysis: Joinpoint regression modeling was applied to identify trends and the Average Annual Percent Change (AAPC) was calculated with 95% confidence intervals and 5% significance level; (3) spatial analysis: spatial dependence among municipalities was assessed using the Global Moran's I index, followed by spatial scan statistics to identify spatial clusters of virus transmission, using both purely spatial and spatiotemporal analyses. A total of 176,122 cases were reported during the study period. Among them, 42.27% (n = 74,458) were aged 20-39 years, 66.96% (n = 117,945) were female, and 34.92% (n = 61,513) were mixed-race. A decreasing trend in incidence rates was observed in Brazil (AAPC - 36.55%; p < 0.001), with disparities across states. Fifteen out of 27 states showed a downward trend. Rio Grande do Norte was the only state to exhibit an increasing incidence trend and was also identified as the area with the highest relative risk of Zika infection in the country (RR = 12.1; p < 0.001). In the spatio-temporal analysis, the cluster with the highest relative risk (RR 31.08; p < 0.001) included 2,770 municipalities across the North, Northeast, Southeast, and Central-West regions. Zika cases predominated among adults, females, and mixed-race individuals. In Brazil, the overall incidence rate showed a decreasing trend during the study period. The Northeast and Central-West regions had the largest number of clusters, highlighting the need to strengthen surveillance activities in these areas to reduce transmission rates.
Status epilepticus (SE) is a neurological emergency associated with high morbidity and mortality. Benzodiazepines (BZDs) are recommended as first-line treatment within minutes of seizure onset; however, treatment delays remain common, particularly in low- and middle-income countries, where data regarding determinants of refractory status epilepticus (RSE) are limited. We conducted a retrospective single-center cohort study including consecutive patients aged ≥ 16 years with SE. The primary objective was to identify factors associated with progression to RSE, with particular focus on latency to first BZD administration. Multivariable logistic regression analysis was performed. A total of 109 SE episodes were analyzed, of which 62 (56.9 %) progressed to RSE. Early BZD administration (≤30 min) occurred in 34.9 % of cases, whereas latency was undocumented in 14.7 %. Early treatment was more frequent in non-RSE than RSE episodes (48.9 % vs 24.2 %), while unknown latency was more common in RSE cases than non-RSE (22.6 % vs 4.3 %; p < 0.01). Two multivariable models were built, differing in how time to BZD was coded. In Model 1 (three-level time-to-BZD variable), undocumented latency was associated with RSE (OR = 9.3, 95 %CI:1.77-49.0; p = 0.01); administration beyond 60 min (OR = 2.4, 95 % CI:0.91-6.63; p = 0.09) and acute symptomatic etiology (OR = 2.2, 95 %CI:0.96-5.0; p = 0.06) were not. In Model 2 (dichotomized time-to-BZD), acute symptomatic etiology was independently associated with RSE (OR = 2.3, 95 %CI:1.05-5.22; p = 0.04). Undocumented latency to BZD administration and acute symptomatic etiology were independently associated with RSE, highlighting challenges of timely SE recognition and supporting efforts to standardize treatment pathways in resource-limited settings.
While scientific environments have been described as unwelcoming to the LGBTQ+ community, and fields such as physics have systematically documented these challenges, the climate in biology-specific workplaces has not exclusively been assessed. We conducted the largest survey to date of LGBTQ+ biologists to examine how their sense of belonging and perception of climate in the biology workplace and professional societies compare to that of their straight and cis peers. In 2023, we surveyed 1419 biologists across five professional societies, with 486 identifying as LGBTQ+. Trans and gender non-conforming (TGNC) biologists reported lower belonging and morale within the workplace, professional societies, and the biology community compared with cis, straight biologists. They also reported being less comfortable with the climate of various professional biology environments. While LGBTQ+ biologists report that their workplaces are moderately inclusive, over 20% of all LGBTQ+ biologists and nearly 40% of TGNC biologists experience exclusionary behavior at work. This landmark survey provides the first comprehensive analysis of the LGBTQ+ climate in biology, revealing specific challenges faced by TGNC scientists.
Ameroid ring constrictors (ARC) are used in veterinary medicine to provide gradual occlusion of portosystemic shunts. To date, ARCs with an inner lumen diameter of 3.5-mm are the smallest that have been studied. Smaller sizes are available but there is evidence from previous studies to indicate that suggest closure dynamics may vary with ARC size. This study compared the closure dynamics of smaller ARCs with previously evaluated larger sizes. Twenty four ARCs, six of each size: 2.5-mm, 3.0-mm, 3.5-mm and 5.0-mm were incubated in canine plasma for 28 days and were digitally imaged prior to the start of study and then on days 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, and 28. The inner lumen diameter was compared for rate and percent of closure over time. No ARC evaluated had completely closed by 28 days. Smaller ring sizes closed faster and to a greater extent than larger ring sizes. There was a significant difference in time to achieve 50% closure with post hoc testing showing that time to achieve 50% closure was faster for 2.5-mm and 3.0-mm rings vs 3.5-mm rings and 3.0-mm vs 5.0-mm rings. In conclusion, the rate and proportion of lumen closure was greater with ARCs with smaller initial lumen size. Clinicians should be aware that use of these smaller sized ARCs will likely result in a more rapid and more complete closure in comparison to larger size constrictors.
Many patients in the United States live with left ventricular assist devices - surgically implanted mechanical pumps that support circulation in individuals with advanced heart failure. The use of these devices has grown steadily, with new implants performed each year and many patients maintained with ongoing longitudinal care. As outcomes improve and patients live longer with this form of mechanical support, clinicians across disciplines are increasingly confronted with the complex clinical, ethical, and practical challenges that arise when life-sustaining therapy is dependent on implanted technology, particularly in the context of end-of-life decision-making. We present a case report of a 76-year-old man with advanced heart failure supported by an implanted cardiac pump. He pursued medical aid in dying - the legal process by which a terminally ill, mentally capable adult may take prescribed medications to end their life. Deactivating a cardiac assist pump typically occurs in a hospital with intravenous sedation for the abrupt heart failure symptoms that can occur when the pump is turned off. But this patient desired to die at home. An end-of-life care navigator and a carefully assembled interdisciplinary team providing medical aid in dying in conjunction with pump deactivation achieved the patient's home death at his selected date. Coordinating implanted cardiac device withdrawal with medical aid in dying is clinically achievable, ethically defensible, and legally sound. Healthcare systems must develop written protocols, train hospice providers, provide anticipatory counseling, fund necessary infrastructure, and support this end-of-life care.
暂无摘要（点击查看详情）
暂无摘要（点击查看详情）