Category
449 papers
External foundation-model validation of MSI counterfactuals
Abstract. Aerosol particles, as crucial atmospheric components, significantly influence optical properties and play an important role in climate change. Based on Lorenz-Mie's theory, the optical equivalent radius of aerosol particle affects the atmosphere visibility. Traditional aerosol models, which assume spherical or other geometrically regular particle shapes, significantly deviate from observations in visibility simulations. In this study, the mechanism of random diffusion and aggregation of monomeric particles into fractal geometrical clusters was reproduced. The Fractal Aerosol Cluster Model (FACM) was developed to parameterize the optical and aerodynamic sizes of the fractal geometry of aerosol particles. Sensitivity experiments were conducted to simulate a severe haze event in northern China in November 2018 by coupling FACM into WRF-Chem as the experimental case (EXP) while the control case (CTR) by the original WRF-Chem. The simulated near-ground PM2.5 concentrations in both EXP and CTR are similar to the observations (OBS). However, EXP simulated the larger extinction coefficients and lower atmospheric visibility (AV), which is more closely to OBS. The average normalized mean error of AV by EXP to OBS is 163.39 %, compared to 421.62 % by CTR. Thus, considering the fractal geometry of aerosol particles significantly improves simulated AV. Furthermore, a reduction of approximately 60 W·m⁻² in land surface shortwave radiation in EXP than those by CTR was also confirmed by observations. This study of the optical properties of the fractal aerosol cluster will contribute to future research of atmospheric environment and climate change forcing.
nonMSIH to MSIH counterfactuals with cell segmentation
Background/Objectives: Peritoneal dissemination of malignancies leads to poor prognoses, and no effective treatment currently exists. The difficulty of treating such malignancies is likely because systemically administered drugs cannot easily target malignant cells in the abdominal cavity. High intraperitoneal drug retention, non-toxicity towards normal tissues, and successful targeting of malignant cells are important for an effective therapy. The aim of this study was to evaluate an intraperitoneally administered astatine-labeled, integrin-targeted nanodrug, mPEG(Mn:350)-S-AuNP[211At]-c[RGDfK(C)] ([211At]AuNP@PEG/RGD), with respect to its kinetics, therapeutic efficacy, and safety. Methods: C6 rat glioma cells (107), and BxPC3 (107) and PANC-1 (107) human pancreatic cancer cells were seeded intraperitoneally into nude mice, and [211At]AuNP@PEG/RGD (0.979 ± 0.194 MBq for C6 models (n = 3), 1.139 ± 0.035 MBq for BxPC3 models (n = 10), and 1.308 ± 0.039 MBq for PANC-1 models (n = 10) per mouse) or saline were intraperitoneally administered 4–7 days later. Cytotoxicity against malignant cells, pharmacokinetics after administration, therapeutic efficacy, and safety in abdominal organs were evaluated. Results: Intraperitoneally administered [211At]AuNP@PEG/RGD accumulated exclusively in the peritoneal cavity for a long period of time and showed minimal systemic diffusion through the blood. In the C6 model, the intraperitoneal tumor mass was significantly lower in the treated group compared with that of the controls (p = 0.05). For the BxPC3 (median survival time: control/treated = 41/65 days, p < 0.001) and PANC-1 (median survival time: control/treated = 19/35 days, p < 0.001) peritoneal dissemination models, survival analysis revealed that [211At]AuNP@PEG/RGD significantly prolonged overall survival. Although transient weight loss, leukopenia, and thrombocytopenia were observed at one week post-administration, a short recovery trend was evident thereafter. One month after administration, no abnormalities were found in hematological tests or histological analyses of intra-abdominal organs. Conclusions: The intraperitoneal administration of astatine-labeled integrin-targeted [211At]AuNP@PEG/RGD nanoparticles showed promising findings in terms of safety and efficacy for treating peritoneally disseminated malignant tumors.
Reach audiences
Advertise in front of researchers, engineers, and readers.
Background/Objectives: Non-contiguous dual spinal cord injuries are associated with poor neurological outcomes, probably due to reduced spared-tract bridges for compensatory reorganization. Intravenous (IV) infusion of mesenchymal stem cells (MSCs) promotes recovery after SCI. However, how MSCs influence long-tract reorganization across multiple lesions remains unclear. We characterized MSC-induced long-tract plasticity at two separate lesion levels using ex vivo diffusion tensor imaging (DTI) in a rat staggered hemisection model. Methods: Adult rats underwent staggered lateral hemisection at right T7 and left T12 and received IV MSCs (1 × 106) or vehicle on day 1; locomotor recovery was assessed over 4 weeks (using Basso–Beattie–Bresnahan (BBB) scale) and ex vivo DTI tractography was performed at 4 weeks (7.0 T). Results: MSC-treated rats showed significantly greater locomotor recovery than vehicle-treated rats (BBB at day 28:12.2 ± 0.6 vs. 5.4 ± 0.7, p < 0.001). MSC infusion increased total, preserved, and redirected tract counts at both lesion levels; at T7, redirected tracts entered the non-injured lateral funiculus, forming detour pathways bypassing the lesion, whereas at T12, redirected tracts from the injured-side dorsal funiculus crossed the midline into the contralateral dorsal funiculus. Conclusions: Intravenous infusion of MSCs improved locomotor outcome and altered the level-specific patterns of long tracts reconstructed from DTI data. Ex vivo DTI provides an overview of these structural changes.
Counterfactual image examples generated for the liver cancer type classifier [hepatocellular carcinoma (HCC) vs. cholangiocarcinoma (CCA)]. A, Representative examples of counterfactual transitions generated with two approaches (linear and MIL) and a comparison of corresponding segmentation masks. B, Counterfactual image generation effectiveness, measured as the percentage of generated images predicted as the opposite class across varying manipulation amplitudes. HCC-to-CCA required higher amplitudes to achieve similar results, which likely reflects class imbalance in the dataset. Scale bar applies to all images within the panel.
Independent classifier validation and representative bidirectional counterfactual transformations in a multiclass setting. A, External classifier responses to counterfactual morphing. Counterfactual images were generated at increasing morphing amplitudes (α) with MoPaDi and then encoded with three foundation models (UNI2, CONCH, and Virchow2). Independent classifiers were trained on the corresponding encoders’ features extracted from all TCGA-CRC tiles and then used to predict the target probability P(target) on both the original and counterfactual tiles. The resulting change ΔP(target) reflects how strongly the morphing affected class evidence. Bars show the median ΔP(target) with percentile-based variability across tiles. B, Representative examples of bidirectional counterfactual explanations for MSIL patients. We defined MSIL by fitting a two-component Gaussian mixture model to the log-transformed distribution of total MSI events and using the intersection of the two components as the cutoff separating MSIL from MSIH samples.
Transition-metal modification of cerium dioxide (CeO2) is widely employed to enhance catalytic performance through the creation of active sites, promotion of oxygen-vacancy formation, and improved oxygen mobility. All of these effects can be strongly influenced by the dynamic redistribution of dopant species under reaction conditions, a phenomenon that remains poorly understood. Here, we investigate the redox-dependent redistributing of Fe in a well-defined FeOx/CeO2(111) model systems using scanning tunneling microscopy, near-ambient pressure X-ray photoelectron spectroscopy, low-energy electron diffraction, and ion sputtering depth profiling by introducing ultra-high vacuum conditions and reducing (H2) or oxidizing (O2) environments at temperatures up to 850 K. Under reducing conditions, the FeOx overlayer becomes unstable above 600–750 K and undergoes extensive disintegration accompanied by Fe diffusion into the CeO2 bulk. Depth profiling reveals a non-uniform Fe distribution with preferential accumulation in deeper regions of the ceria film, eventually approaching the CeO2(111)/ Pt(111) interface. In contrast, oxidizing conditions stabilize Fe at the CeO2(111) surface over the entire investigated temperature range accompanied by substantial restructuring of FeOx species on the surface. Fe incorporation is shown to be reversible, as oxygen annealing of samples with Fe-depleted surfaces restores Fe to the surface. These results demonstrate that FeOx/CeO2(111) systems exhibit fundamentally different structural and chemical behavior under reducing and oxidizing environments, enabling a dynamic redistribution of Fe. The observed activation temperatures for Fe redistribution closely coincide with those reported for Fe–CeO2 catalyst activation in hydrogen temperature-programmed reduction experiments, suggesting that redox-controlled Fe redistribution represent a key structural process governing the performance of Fe-containing ceria catalysts.
Tissue-type counterfactual examples for additional classes
Despite growing global interest in Artificial Intelligence (AI) in education, limited empirical research has examined how teacher education institutions in developing-country and resource-constrained contexts can sustainably integrate AI into their curricula. This gap is particularly significant because many existing AI integration frameworks assume levels of technological infrastructure, institutional capacity, and digital readiness that may not reflect the realities of higher education institutions in the Global South. Against this backdrop, this study explored institutional readiness, curriculum integration pathways, and implementation strategies for Artificial Intelligence within teacher education programmes in Namibia. Guided by an interpretivist paradigm, a qualitative exploratory design was employed, using semi-structured interviews with 25 teacher educators across six university campuses. Data were analysed thematically using Braun and Clarke’s framework. The findings indicate that AI integration extends beyond technological adoption and requires coordinated curriculum transformation, institutional preparedness, and pedagogical redesign. Key integration pathways include embedding AI within faculty courses, curriculum-wide integration of AI competencies, simulation-based learning, and support for online teaching environments. Institutional readiness, particularly in terms of infrastructure, faculty capacity, and curriculum alignment, emerged as a critical determinant of implementation. While participants highlighted benefits such as improved instructional efficiency, enhanced teacher capacity, and increased learner autonomy, they also expressed concerns regarding overreliance on technology. By integrating Diffusion of Innovation (DOI) and the Technology Acceptance Model (TAM), this study advances a multi-level framework linking institutional and individual dimensions of AI adoption in teacher education. This study contributes context-specific insights to AI curriculum transformation in the Global South and provides practical implications for curriculum design, institutional strategy, and policy development.
The accuracy of the Atlung Method for Intercalant Diffusion and Resistance (AMIDR) for determining solid-state lithium diffusion was evaluated using synthetic data generated with the Single Particle Model (SPM) and Doyle–Fuller–Newman (DFN) models implemented in PyBaMM. Under SPM conditions, the AMIDR model achieved single-digit errors across a wide range of parameters but exhibited significant fitting errors at extreme parameter values. DFN simulations revealed an additional systematic underestimation of diffusion at high DNMC due to electrolyte transport, with errors approaching two orders of magnitude despite optimized cell design.
Data Management
What if pathology foundation models could do more with less? GigaPath-Flash and GigaTIME-Flash cut computational demands while maintaining strong performance, opening the door to larger studies and broader exploration. The post GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models appeared first on Microsoft Research.
