Large-scale acoustic simulation is essential for ultrasound imaging and therapy but faces computational bottlenecks when modeling nonlinearity, frequency-dependent attenuation, and absorbing boundaries in heterogeneous media. To address this, we present Fullwave 2, a unified time-domain formulation that integrates power-law tissue attenuation and Perfectly Matched Layers (PMLs) via complex coordinate stretching. This framework preserves the structure of the d'Alembertian operator, enabling the use of high-order staggered-grid finite difference stencils optimized for long distance propagation. Since the attenuation in the interior of the domain uses the same formulation as the boundaries, sophisticated implementations of the PMLs are easily implementable and do not add any computational burden. Here a two-stage Convolutional Perfectly Matched Layer (C-PML) with a transition region is demonstrated to be highly stable. The PML attains high absorption efficiency, achieving reflection coefficients below -50dB with a compact $4\lambda$ footprint. The domain-wide multiple relaxation mechanisms achieve <5% attenuation error and <0.5% phase-velocity error over a 1--20 MHz bandwidth. Validation against a 1D Burgers solution confirms nonlinear accuracy up to the 7th harmonic. The capabilities of the simulation tool are finally demonstrated in ultrasound imaging of 2D abdominal phantoms, and in 3D transcranial applications demonstrating how the method accurately captures complex scattering and aberration artifacts, and provides an efficient and acoustical wave propagation tool for medical ultrasound research.
This randomized trial was designed to compare 1-yearoutcomes in patients who received revascularization based on computed tomography-derived fractional flow reserve(CT-FFR) versus quantitative coronary angiography(QCA). In this early analysis, early angiographic and clinical outcomes were compared. A total of 106 patients were randomized to undergo coronary artery bypass grafting based on CT-FFR(CT-FFR group,N=53) or QCA(QCA group,N=53). Complete revascularization was achieved for coronary arteries with CT-FFR ≤0.78(CT-FFR group) or those with QCA stenosis(>70% stenosis for left; ≥90% stenosis for right coronary artery territories)(QCA group).. Angiography was performed at median 1[1,1] postoperative day in all patients. Angiographic findings of graft flow were categorized as perfectly patent, competitive(bidirectionally or unidirectionally competitive), or occluded. There were no differences in patient characteristics between the 2 groups. The average number of revascularized coronary arteries per patient were 3.7±0.8 in the CT-FFR group and 3.5±0.7 in the QCA group(P=0.089). The median difference in expected anastomosis sites was 2[1,3]. There were no operative mortality and no differences in morbidities between the 2 groups. The number of perfectly patent, bidirectionally competitive, and unidirectionally competitive grafts were 167(85.2%),10(5.1%), and 19(9.7%),respectively, in the CT-FFR group(n=196);133(72.7%),23(12.6%), and 27(14.8%),respectively, in the QCA group(n=183). There was a significant difference in the number of perfectly patent grafts between the 2 groups(p=0.007). The average number of distal anastomoses was similar between the two groups. However, the number of competitive grafts was significantly reduced in the CT-FFR group compared to the QCA group.
Peptide nucleic acid (PNA) offers superior chemical stability relative to DNA, enabling encoded library synthesis under conditions incompatible with conventional DNA-based systems. Here, we report a robust PNA encoding strategy for the construction of large single-pharmacophore PNA-encoded libraries through enhanced discrimination between perfectly matched PNA/DNA duplexes and mismatches. This approach integrates a mismatch-destabilizing codon design with optimized hybridization conditions that selectively favor formation of perfectly matched duplexes.
We study a simple model for a particle that is active due to repulsive self-phoresis and that has been proposed to model symmetric camphor grains. The particle generates a concentration field through the continuous emission of a chemical substance and its motion is driven by gradients of this field as it diffuses within a confined channel whose ends perfectly reflect the chemical. The reflection of the chemical field leads to an effective confinement of the particle, which itself is reflected before encountering the channel ends. The system displays a transition from a passive state, where the particle rests at the channel midpoint, to an active state characterized by highly regular, non-chaotic oscillations. We analytically construct the phase diagram and derive the oscillation frequency and amplitude in the vicinity of the transition. A perturbative analysis perfectly describes the dynamics of the particle, even for oscillations as large as half the channel size. Furthermore, we develop an analysis which explains the mechanism of particle reflection close to the channel edges in the regime of large activity.
Prion diseases can mimic Alzheimer disease (AD) at presentation. Alzheimer's Association AD diagnostic criteria suggest that a single abnormal highly specific plasma biomarker (including p-tau217) is sufficient for a biological diagnosis. We investigated the performance of AD plasma biomarkers in distinguishing AD and prion diseases. We examined plasma biomarker data from patients with prion disease from a prospective cohort study recruited through the UK National Prion Clinic. Prion diseases were diagnosed clinically or with autopsy confirmation, and AD was diagnosed clinically with CSF biomarker confirmation. Plasma p-tau217, p-tau181, Aβ42/40 ratio, brain-derived tau (BD-tau), neurofilament light chain (NfL), and glial fibrillary acid protein (GFAP) were measured using Simoa. Median biomarker values in different groups were compared with Kruskal-Wallis test, and area under the receiver operating characteristic curve was used to compare accuracy in distinguishing prion diseases from sporadic AD (sAD). Lumipulse p-tau217 and NfL were measured in a validation study in a different laboratory. In the main study, we analyzed 345 samples from 278 individuals (mean age 58 [SD 13.5], 48.2% female), including 204 with prion diseases (121 sporadic Creutzfeldt-Jakob disease [CJD], 11 iatrogenic CJD, 9 variant CJD, 47 slow-progressing inherited prion disease (IPD) and 16 fast-progressing IPD), 33 with AD, and 41 healthy controls. For discriminating prion disease without AD copathology from sAD, none of p-tau217 (area under the curve [AUC] [95% CI] 0.605 [0.486-0.724]), p-tau181 (AUC 0.554 [0.446-0.661]), or GFAP (AUC 0.514 [0.389-0.640]) performed well. Aβ42/40 discriminated moderately (AUC 0.770 [0.684-0.856]). NfL/p-tau217 ratio (AUC 0.996 [0.987-1.000]), NfL (AUC 0.988 [0.974-1.000]), BD-tau/p-tau217 ratio (AUC 0.963 [0.929-0.996]), and BD-tau (AUC 0.934 [0.890-0.978]) discriminated very well. In an independent validation study, consecutive samples were analyzed from 32 patients with sAD and 35 patients with sporadic Creutzfeldt-Jakob disease (mean age 65.0 [SD 6.4], 56.7% female). NfL/p-tau217 again discriminated almost perfectly (AUC 0.986 [95% CI 0.966-1.000]). Plasma p-tau217 and p-tau181 are increased in both AD and prion diseases (regardless of burden of AD copathology). Diagnosing AD with a single abnormal p-tau plasma biomarker risks misdiagnosing prion diseases as AD. Plasma NfL/p-tau217 discriminates near-perfectly and could act as a flag to suspect prion diseases where this is a diagnostic possibility. This study provides Class II evidence that plasma NfL/p-tau217 discriminates patients with CJD from those with AD.
Medical in-kind benefits significantly influence households' real disposable income and medical consumption capacity. However, traditional household survey data typically lack information on in-kind benefits. To directly measure the distribution of medical in-kind benefits and to accurately estimate their effects on the inequality of income and medical consumption, this study linked income distribution data from Statistics Korea's Survey of Household Finances and Living Conditions (SHFLC) with medical in-kind benefits administrative data sourced from the National Health Insurance Service (NHIS). While an 80 percent match rate was achieved at the individual level, only 53 percent of the SHFLC samples were perfectly matched at the household level, with imperfect matches more prevalent in larger households. Our preliminary analysis, based on reweighted perfectly matched samples, yielded several key findings: The medical in-kind benefits-including health insurance, long-term care, and social assistance-were highly progressively distributed by income, with average benefits being higher for low-income households and the elderly. Aggregately, benefits provided to the bottom income quintile were more than three times those provided to the top. The inclusion of these benefits reduced poverty rates based on adjusted disposable income by approximately 2 percentage points for the entire population and by about 10 percentage points for the elderly. These initial findings robustly demonstrate the significant redistributive power of medical in-kind benefits in mitigating poverty among vulnerable populations. To address potential selection bias from imperfect matching, we will conduct robustness checks using multiple imputation techniques for the missing data.
Competitive graft flow caused by functionally non-significant coronary stenosis may compromise graft performance after coronary artery bypass grafting (CABG). We compared the ability of preoperative computed tomography-derived fractional flow reserve (CT-FFR) and intraoperative transit-time flow measurement (TTFM) to predict competitive graft flow on early postoperative angiography. This retrospective study included 235 patients who underwent isolated CABG using an in situ internal thoracic artery-based composite graft and for whom preoperative CT-FFR values were available. Intraoperative TTFM parameters, including mean graft flow (MGF), pulsatility index (PI), diastolic filling percentage (DF%), and percentage of backward flow (%BF), were recorded for each anastomosis. Early postoperative angiography was performed in all patients, and graft flow was classified as perfectly patent, bidirectionally competitive, unidirectionally competitive, or occluded. Receiver operating characteristic analyses were performed to compare the predictive performance of CT-FFR and TTFM parameters. Among 744 anastomoses (median, 3 [3.0, 4.0] per patient), 602 (80.9%) were perfectly patent, 69 (9.3%) were bidirectionally competitive, 70 (9.4%) were unidirectionally competitive, and 3 (0.4%) were occluded. Computed tomography-derived fractional flow reserve values increased according to the degree of competitive graft flow and demonstrated the highest predictive accuracy for competitive flow (cutoff, 0.764; area under the curve [AUC], 0.845; P < .001). The predictive performance of CT-FFR was superior to that of all TTFM parameters, including MGF (AUC, 0.730), PI (AUC, 0.701), DF%(AUC, 0.625), and %BF (AUC, 0.707) (all P < .001). In patients undergoing composite CABG, preoperative CT-FFR demonstrated superior predictive performance compared with intraoperative TTFM parameters for early competitive graft flow. These findings support the integration of physiologic lesion assessment into surgical planning to optimize revascularization strategy.
To compare the effectiveness of closed-loop (AutoStim) versus open-loop vagus nerve stimulation (VNS) in drug-resistant epilepsy (DRE), and to explore predictors of treatment response. We conducted a single-centre retrospective cohort study at Xuanwu Hospital (2012-2024). A total of 131 consecutive DRE patients received open-loop (Model 102) or closed-loop (Model 106) VNS; 81 were analysed after follow-up (open-loop n = 44; closed-loop n = 37). The primary outcome was percent reduction in monthly seizure frequency; responders were defined as ≥ 50% reduction. Between-group comparisons accounted for baseline age imbalance using Quade rank ANCOVA. Predictor analyses evaluated clinical covariates, including age at seizure onset. Closed-loop VNS was associated with greater postoperative seizure reduction than open-loop after age adjustment (Quade rank ANCOVA p = 0.007). Overall responder rate was 58% (47/81); by group: 43.2% (19/44) in open-loop versus 75.7% (28/37) in closed-loop, absolute difference 33.8 %age points (95% CI 8.9-58.7), NNT= 3 (95% CI 2-3). Age at seizure onset was associated with response in exploratory analyses (t(79) = -2.241, p = 0.028). In an Asian cohort, closed-loop VNS seems to be associated with greater seizure reduction than open-loop VNS in this real-world cohort; however, because device type was perfectly aligned with implantation era, findings should be interpreted as comparative effectiveness across eras rather than definitive causal superiority. Age at seizure onset shows an exploratory association with response, but this should be interpreted cautiously given differential attrition and era-related confounding.
Owing to the existence of gravel of tight sand conglomerate, the law of propagation and main factors of hydraulic fracture have not been understood perfectly. Firstly, the finite-discrete element method (FDEM) was used to investigate the ductile behavior of hydraulic fractures when they met gravel. Then, the main geological and engineering factors affecting hydraulic fracture propagation are determined, such as difference of horizontal principal stress, gravel size, viscosity of fracturing fluid and displacement. Subsequently, the boundary element method (BEM) was used to demonstrate that the propagation process of macroscopic artificial fractures in conglomerate can be simulated by equivalent natural fractures. The results indicated that: (1) The fracture morphology of sand conglomerate reservoir was determined by stress state and gravel, mainly dominated by gravel-bypassing and bifurcation under low stress difference; (2) The spacing of natural fracture has the greatest positive influence on the length of hydraulic fracture, while the minimum horizontal principal stress has the greatest negative influence; (3) The smaller the angle of natural fracture is, the greater the frequency of branching fracture occurs; 4)The hydraulic fractures in tight conglomerate reservoir are mainly characterized by bi-wing fractures on the macro levels. The results can provide some reference for revealing the propagation mechanism of hydraulic fracture and selection of fracturing section in conglomerate reservoir.
Polyethylene represents an exceptional thermal conductor in theory: The perfectly extended chain is predicted to conduct heat extremely well. However, its practical scalable forms fall far below this limit because noncrystalline structures disrupt heat transport across multiple length scales. Here, we identify that a partially ordered, noncrystalline transitional phase in highly aligned polyethylene is not a negative by-product of processing but a key contributor to heat conduction. Guided by this insight, we develop a gel-state intermittent slow stretching method that directs noncrystalline evolution during polyethylene fiber formation. This approach enables chain relaxation and structural reorganization in regions typically regarded as amorphous and interfacial, promoting their conversion into the transitional phase and their seamless integration with crystalline domains. The resulting structure extends the continuity of ordered shish segments within the period structure, increasing the distance over which heat can travel quasi-ballistically. As a result, polyethylene fibers with 26.79% noncrystalline content achieve thermal conductivities up to 70.61 watts per meter per kelvin, representing 1.45 to 2.60 times of leading commercial polyethylene fibers. These findings establish control of noncrystalline structure as an essential route to unlocking high thermal conductivity in polymers and open a pathway toward lightweight, fully organic thermal conductors.
The capacity of phenylalanine homopeptides to form nanoscale fibrils holds great potential in the fields of materials science, biomedicine, and nanotechnology. Yet, detailed structures are still missing due to their inherent inability to form perfectly ordered lattices. Here, using multiscale molecular dynamics simulations, we describe the spontaneous assembly pathway of tetraphenylalanine (Phe4) peptides, revealing a coaxial fibril architecture with a trigonal cross-β motif at pseudo-atomic resolution. Starting from isolated peptides in solution, multi-microsecond coarse-grained simulations of hundreds of Phe4 molecules capture hierarchical fibril formation without structural bias. An idealized fibril model built from these results was validated against experimental X-ray powder diffraction data, showing strong agreement with observed lattice parameters, and all-atom molecular dynamics simulations supported its stability. Guided by simulation-derived insights into the kinetics of longitudinal versus lateral growth, we further demonstrate that unbranched Phe4 fibrils, the smallest reported to date, can be selectively obtained by harvesting supernatant fractions from low-concentration assemblies, as confirmed by transmission electron microscopy. Beyond the structural characterization of Phe4, the ab initio coarse-grained seeding protocol introduced here represents a broadly applicable strategy for modeling early aggregation states of amyloidogenic peptides.
Modern neuroprostheses can now restore communication in patients who have lost the ability to speak or move. However, implanting these invasive devices comes with risks inherent to neurosurgery. Here we introduce a noninvasive method to decode the production of sentences from brain activity and demonstrate its efficacy in a cohort of 35 healthy volunteers. For this, we present Brain2Qwerty, a new deep learning architecture trained to decode sentences from either electro- or magnetoencephalography, while participants typed briefly memorized sentences on a QWERTY keyboard. With magnetoencephalography, Brain2Qwerty reaches, on average, a character error rate of 29% and substantially outperforms electroencephalography (character error rate: 65%). For the best participants, the model achieves a character error rate of 18%, and can perfectly decode a variety of sentences outside of the training set. Overall, these results narrow the gap between invasive and noninvasive methods and thus open the path for developing safe brain-computer interfaces for noncommunicating patients.
Introductory biology instructors contend with the tension between the inherent complexity of the discipline and cognitively overloading their students. While oversimplification may promote misconceptions, without sufficient simplification, students may fail to grasp a basic understanding of disciplinary concepts. In genetics education, evidence is mounting that the risks associated with oversimplifying far outweigh the benefits, as emphasizing Mendelian principles while neglecting the role of environment not only fails to reflect our current understanding of genes and genomes but can also inadvertently reinforce the misinformed belief that genes alone determine people's physical, cognitive, and behavioral characteristics. Examples and activities featuring human pedigrees can be particularly insidious. While excellent tools for helping students apply fundamental genetics concepts and practice analytical skills, they reduce phenotypic variation to a dichotomous "affected" vs. "unaffected" model with perfectly Mendelian inheritance and the underlying, unspoken assumption that phenotype is entirely determined by genotype. There is a need for instructional resources that present genetics in a way that better reflects present-day knowledge, make light of the complexities hidden behind pedigree charts, and allow students to appreciate the multifactorial nature of phenotypic variation, and the challenges of classifying individuals into discrete categories. To help fill this gap, we offer an engaging lesson that promotes critical pedigree analysis skills and honors the complexities of phenotypic variation.
High numerical aperture objective focusing is essential for resolving subwavelength structures. However, it induces significant transverse polarization crosstalk in Mueller matrix microscopy (MMM). To address this issue, we present an analytical framework to achieve deterministic spatial polarization preservation. By inverting the vectorial diffraction tensor, we prescribe an incident complex amplitude that annihilates orthogonal crosstalk via destructive interference. This active pupil modulation ensures that arbitrary input polarization states are uniformly maintained across the focal plane. Additionally, we develop a complementary algorithm to rectify the secondary distortion arising from the collection objective, which forcibly projects the intrinsic longitudinal field and generates a coherent parasitic background. By cascading both correction strategies, all Mueller matrix elements can be perfectly restored. These findings not only enable distortion-free MMM but also establish a reliable foundation for polarization-sensitive nanofabrication and multidimensional optical storage.
To assess the risk of stroke/TIA and progression to atrial fibrillation (AF) in Indian patients with device-detected atrial high-rate episodes (AHRE) lasting less than 24 h. In this single-centre prospective matched cohort study, 109 patients with AHRE < 24 h matched to 109 controls without AHRE on age (+/-5 years), sex, device type, device indication, and CHA2DS2-VASc score. Followed for a minimum of 12 months (mean 18.4 months). The CHA2DS2-VASc score was near-perfectly matched between groups (mean 3.26 vs. 3.29, score distribution chi-square p = 1.00). A stroke or TIA occurred in 7 patients with AHRE versus 1 control (annual rate 4.28% vs. 0.61%; OR 7.41, 95% CI 0.90-61.3; p = 0.020; HR 7.19, 95% CI 0.89-58.48; log-rank p = 0.031). Progression to clinical AF, AHRE ≥ 24 h, atrial flutter, or atrial tachycardia happened in 32 AHRE patients versus 2 controls (annual rate 19.7% vs. 1.2%; OR 22.2, 95% CI 5.17-95.6; HR 19.55, 95% CI 4.6-81.71; log rank p < 0.001). Stroke risk increased with CHA2DS2-VASc score: 2.15% (score 3), 4.16% (score 4), 6.66% (score 5), and 9.5% (score ≥ 6). In this hypothesis-generating study, device-detected AF lasting less than 24 h is associated with a significantly higher risk of stroke/TIA and progression to clinical AF in Indian patients. It requires confirmation in larger studies. Region-specific absolute risks must be considered when applying international anticoagulation guidelines.
Preoperative characterization of pancreatic cystic lesions (PCLs) remains inaccurate when based on morphology alone. We aimed to determine whether integrating complementary imaging (cross-sectional imaging plus endoscopic ultrasound) and clinical history, particularly history of pancreatitis, improves diagnostic accuracy compared with single-modality imaging. This retrospective study included consecutive patients with PCLs treated at a tertiary referral center (test cohort) and a confirmatory cohort from a second hospital. Preoperative diagnostic accuracy was assessed in patients with histological confirmation. Multivariable logistic regression identified independent predictors of correct preoperative diagnosis. Additionally, a secondary multivariable logistic regression analysis included all patients using final diagnosis at follow-up (mean 2.4 ± 0.3 years) was performed. In the test cohort, 258 patients were analyzed; 59 (23%) had histological confirmation. Overall preoperative accuracy was 63% (validation cohort: 68%). No significant difference was observed between MRI, CT, or endoscopic ultrasound alone. Complementary imaging was independently associated with higher diagnostic accuracy in multivariable analysis, as was main pancreatic duct dilation. In the full cohort, a history of acute or chronic pancreatitis was the only independent predictor of correct diagnosis and strongly predicted pseudocysts (80% vs. 10% without pancreatitis; p < 0.00001). Combining complementary imaging with pancreatitis history further improved overall diagnostic accuracy and significantly increased detection of malignant PCLs compared with MRI alone (p = 0.03). Diagnostic accuracy of PCLs is limited with single-modality imaging. A stepwise strategy integrating complementary imaging and clinical history, particularly a history of pancreatitis, significantly improves identification of malignant lesions and may refine guideline-based management. Pancreatic cystic lesions are fluid-filled sacs in the pancreas that are increasingly detected because imaging tests such as magnetic resonance imaging and computed tomography are widely used. Most of these cysts are harmless, but some can develop into cancer. Correctly identifying which cysts are dangerous before surgery remains challenging. Doctors often rely on imaging tests to assess these cysts. Another examination method is endoscopic ultrasound, in which a small ultrasound probe is placed inside the stomach to obtain detailed images of the pancreas. However, no single imaging method is perfectly accurate. In this study, we analyzed patients with pancreatic cysts from two hospitals to examine whether combining different imaging methods improves diagnostic accuracy. We also evaluated whether information from the patient’s medical history, especially a previous episode of pancreatitis (inflammation of the pancreas), helps clarify the diagnosis. We found that using more than one imaging method provided more accurate results than using only one test. In addition, a history of pancreatitis strongly increased the likelihood that a cyst was a pseudocyst, which is a benign cyst caused by inflammation rather than cancer. Our findings suggest that a stepwise approach – combining imaging results with relevant medical history – may improve decision-making and help better identify patients who require surgery.
Water, an abundant resource with approximately 1.3 × 1016 liters present in the atmosphere and essential for sustaining life, has long held untapped potential for electromagnetic wave absorption (EWA) due to persistent technical challenges, including difficulties in encapsulation, impedance mismatch, and restriction on a single loss mechanism. To address these limitations, this study develops a multi-phase system via atmospheric water harvesting (AWH), wherein surface tension and capillary action confine inorganic solution to foam surfaces, forming ultrathin liquid films that enable synergistic interaction among the air/water/polymer phases and enhance impedance matching. Experimental results reveal that, within the temperature range of 283-323 K, CaCl2-PI@H2O attains a minimum reflection loss (RL) of -44.6 dB at 4 mm (30% RH - 40% RH), effective absorption bandwidth (EAB) perfectly covers the X band (7.4-12.6 GHz). Whereas LiCl-PI@H2O achieves an RL of -40.4 dB at 4 mm (60% RH -70% RH), EAB remarkably broadens to 8.4-18.0 GHz. The key innovation lies in the incorporation of water into a confined salt-solution film, handling the inherent impedance mismatch and single-loss mechanism limitation of water as an absorbing medium, thereby offering a novel and promising strategy for fabricating high-performance liquid-based EWA materials.
Sphere packings in circular cylinders have attracted substantial research interest, among which the discovery of chiral helical structures is the most iconic. However, recent experimental results on zebrafish do not match the known packing structures in circular cylinders. To account for the inherent imperfections of biological tubes, we take elliptic cylinders as the canonical deformation of circular cylinders and investigate the densest packings of hard spheres in them using simulation, theory, and experiments. Starting from the chiral structures in circular cylinders, we demonstrate that even a weak cross-sectional deformation can trigger entirely new phases, including ones that either eliminate global chirality or significantly complicate the chiral structures. This reveals the significant effect of cylindrical anisotropy. The new helical phases under anisotropic confinement remain chiral and develop hierarchical periodic structures, which are difficult to obtain by simulations but are predicted by our newly developed theory for helical phases in elliptic cylinders. The theory also predicts double oscillated-chain phases without chirality, which perfectly match the simulations. Our work offers fresh insights into understanding packings in anisotropic cylinders, which will help researchers to design new materials and to understand many living systems.
Test-time scaling has emerged as a promising method to enhance the reasoning capabilities of large language models (LLMs) and vision-language models (VLMs) during inference without additional training. While foundational studies established scaling paradigms in general domains, their applicability to the unique complexities of medical AI remains underexplored. This study aims to conduct a comprehensive investigation of test-time scaling in the medical domain. We evaluate the impact of scaling across different model sizes and task complexities. Furthermore, we seek to identify domain-specific bottlenecks and assess model robustness against user-driven perturbations, such as misleading clinical authority. This study evaluated a diverse set of general and medical-specific LLMs and VLMs. Experiments used five textual medical benchmarks comprising over 5500 questions and two multimodal benchmarks comprising 7000 samples. Performance was measured under three scaling conditions: increasing token budgets, iterative sequential scaling, and parallel scaling. Robustness was tested by embedding misleading hints with varying tones and levels of simulated clinical expertise into prompts. For nonreasoning LLMs, accuracy saturated quickly, with token usage often remaining under 500 tokens regardless of budget increases. Reasoning models demonstrated significant performance gains on complex tasks as token budgets increased. Notably, we identified distinct domain-specific behaviors. First, current VLMs showed a structural bottleneck in integrating visual clues and experienced limited benefit from token expansion. Second, medically fine-tuned LLMs excelled in clinical question answering but exhibited degraded scaling efficiency on calculation tasks compared to general-domain models. This reflects a disparity between qualitative clinical alignment and procedural logic. Third, while optimal scaling improved robustness, models exhibited a cognitive vulnerability by readily abandoning correct reasoning when confronted with misleading expert physician hints. Regarding scaling strategies, parallel scaling outperformed sequential scaling on easier tasks. Conversely, extended sequential scaling or increased budgets proved essential for complex problem-solving. Test-time scaling rules from general domains do not perfectly translate to medical AI. Longer reasoning is not universally beneficial. Concise reasoning with parallel scaling is optimal for simpler tasks. An extended chain of thought via sequential scaling or increased budgets is required for complex problems. Furthermore, safe clinical deployment requires addressing fundamental vision-language alignment, balancing clinical and procedural reasoning, and mitigating vulnerabilities to perceived clinical authority.
Autoimmune hepatitis (AIH) is a chronic progressive liver disease that despite suggestive serum autoantibodies or plasma cell enrichment, remains functionally a diagnosis of exclusion. Whether the broader cellular composition of the liver might enable improved specificity of diagnosis has not been systematically tested. We prospectively recruited patients undergoing a clinically-indicated liver biopsy for suspected AIH and performed single-nucleus RNA sequencing (snRNA-seq) on biopsy tissue to map the cellular landscape of AIH and its diagnostic mimics. Unsupervised clustering on cell-type abundances alone largely separated AIH from non-AIH samples. Among individual populations, a subset of CD8⁺ T-cells marked by high TOX and PD1 expression was the most discriminating feature: its enrichment perfectly distinguished AIH by both snRNA-seq and in situ density (AUC = 1.00), outperforming plasma cell abundance (AUC = 0.83). CD8⁺TOX⁺ T-cell enrichment may therefore be the histologic lesion that marks the diagnosis of AIH.