diffusion models

Figure 5 from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

Style–morphology decomposition for disentangling structural and staining effects on MSI prediction changes. A, Representative examples of counterfactual manipulation between MSIH and non-MSIH classes. For each original image x and its counterfactual xcf, style-hybrid (xstyle) and morphology-hybrid (xmorph) images were generated using Vahadane stain transfer. Each hybrid isolates the effect of either stain or morphology while controlling for the other. The right-hand bars show the Shapley-style decomposition of the logit change (Δf) into stain (φstyle) and morphology (φmorph) contributions, demonstrating that morphologic differences dominate the model’s predictions. Grad-CAM visualizations below provide region-level attribution under MIL. In contrast, MoPaDi produces class-directed “what-if” edits that offer a complementary view of candidate morphologic and style changes associated with prediction shifts. Scale bar applies to all images within the panel. B, Decomposition results across test-set patients, showing median contributions of φstyle, φmorph, and total (Δf) for manipulations toward (↑) and away from (↓) each class. C, Scatter plot of morphology versus style contributions per patient, illustrating consistent dominance of morphologic effects across both manipulation directions and classes.

AI-driven radiomics and radiogenomics: supporting the assessment and differentiation of pseudoprogression in cellular immunotherapy for glioblastoma

Glioblastoma (GBM) is the most aggressive type of primary brain tumour in adults, and after treatment, it is also difficult to know whether a person’s health has improved. Pseudoprogression (PsP), particularly following radiotherapy and combination therapy with temozolomide and new immunotherapies, may be falsely identified as true progression (TP) by conventional MRI, thus leading to premature termination of treatment or unnecessary intensification of therapy. Although RANO, iRANO and RANO 2.0 have improved the assessment of response, structural MRI alone is unable to reveal the biological complexity of the tumour microenvironment. Artificial Intelligence (AI)-based radiomics and radiogenomics aid in the characterisation of GBM. Several Magnetic Resonance Imaging (MRI) sequences are used to obtain quantitative data, such as the conventional T1-weighted images, diffusion-weighted images, perfusion-weighted images and others, to acquire information on cell density, blood vessel distribution, immune cell concentration, molecular modifications and the effect of therapy. Combine imaging characteristics with liquid biopsy, genomic data and patient health records to enhance the accuracy of diagnosis and discover high-sensitivity surrogate markers for immune checkpoint inhibitor and CAR-T cell therapy clinical trials. Clinical translation still faces numerous limitations such as inconsistent research protocols, small study cohorts, insufficient external validation, inadequate model interpretability and inconsistent reference standards. In the future, many research groups will conduct multi-centre validation, standardize workflows, open-source reporting and release clinically interpretable models. The above ways can reduce the bias induced by PsP and improve differentiation between pseudoprogression and true progression to facilitate prompt treatment for most people.

Income and density spillover effects in German land prices

Purpose This paper aims to investigate the role of county-level income and population density effects in explaining spatial variation in German land prices from 2014 to 2018. Motivated by recent evidence that land values account for a growing share of housing price dynamics, our analysis pursues three core objectives. First, we quantify the direct and indirect effects of county-level agglomeration variables on land prices. Second, we evaluate the extent to which spillovers in land prices contribute to clustering patterns in major German cities and their suburban counties. Third, we assess whether these effects differ by land type – comparing residential land with vacant land designated for development. Design/methodology/approach Our main empirical model of interest is the Spatial Durbin Model (SDM), which allows for both endogenous spatial dependence in the dependent variable (prices) and exogenous interactions in agglomeration-related covariates (e.g. median income, population density). Findings We find that spatial agglomeration effects, especially in income and density, exert significant local and neighboring effects on land prices. Specifically, a 1% increase in median income leads to a 1.5–2.5% increase in local land prices and an additional 0.5–0.9% increase in neighboring counties. Similarly, a 1% increase in population density raises land prices by 2–3% locally and 1.5–2% indirectly. These effects display greater intensity in suburban counties near large metropolitan areas. Finally, the effects are more pronounced for residential land than vacant land, consistent with the idea that realized use and regulatory constraints magnify spatial externalities in housing markets. Originality/value We contribute to the literature on agglomeration and housing price dynamics by explicitly modeling income- and density-induced spatial spillovers in land prices using a spatial panel framework. This paper thereby provides a link between the literature on spatial housing dynamics, agglomeration variables and land valuation. Our findings suggest that policy interventions targeting housing affordability or regional price disparities should account not only for housing demand and supply but also for the spatial diffusion of land price shocks, especially in high-productivity regions experiencing density-driven growth.

Figure S12 from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

Perceptual realism expert study

Supplementary Tables from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

Supplementary Tables S1-8

A model for the decay of TKE in the CBL from the time evolution of the length and velocity scales: application in a Eulerian dispersion model

Figure S6 from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

Additional tumor to normal colon mucosa counterfactual transitions

Figure S9 from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

Counterfactuals for borderline vs. top-contributing MSI tiles

Figure S2 from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

MIL transformer architecture for patient-level classification

Beyond Traditional Metrics: An Integrative Review and Conceptual Framework for Managing R&D Project Performance in the Petroleum Industry

This review provides an integrative synthesis of the R&D performance measurement literature and introduces the Beyond Traditional Metrics (BTM) framework as a multi-criteria reference model for R&D performance evaluation in the petroleum sector. Research and Development (R&D) constitutes a strategic pillar of the petroleum industry, where technological innovation supports competitiveness, operational efficiency, and the transition toward more sustainable energy systems. However, evaluating the performance of R&D projects remains a major challenge because their outcomes are often uncertain, intangible, long-term, and multidimensional. Commonly used Key Performance Indicators (KPIs)—such as cost, time, and number of deliverables—therefore provide only a partial representation of R&D effectiveness. R&D performance assessment must therefore consider the intrinsic diversity of innovation activities. Reverse engineering emphasizes replication and adaptation of existing technologies, while innovation-driven R&D seeks to create novel knowledge, technological capabilities, and strategic learning. Accordingly, the selection of performance indicators should be adapted according to project type, technological maturity, and strategic objectives. To avoid biased evaluation, the approach integrates principles derived from the Multi-Criteria Decision Analysis (MCDA) approach, enabling prioritization of criteria aligned with each project’s objectives, complexity, and organizational priorities. To move beyond simple cost and time metrics, this study revisits the meaning of “performance” in R&D and explores a multidimensional evaluation perspective capable of capturing both tangible and intangible forms of value creation by integrating five complementary dimensions: Knowledge Creation and Diffusion, Innovation Velocity, Dynamic Strategic Alignment, Team and Organizational Health, and Resilience and Robustness under technological, regulatory, operational, and market uncertainty. The framework is illustrated through an exploratory application to a hypothetical portfolio of petroleum-sector R&D projects, demonstrating its potential usefulness for benchmarking, portfolio prioritization, and multidimensional innovation assessment under conditions of uncertainty.

Figure 7 from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

Counterfactual image examples generated for the lung and breast cancer–type classifiers. A, Representative examples of LUSC tile transitioning to its counterfactual lung adenocarcinoma (LUAD) image and vice versa. B, Morphologic feature prevalence in original and counterfactual image pairs (N = 32 transitions; 16 tiles for each class). Horizontal bars show the percentage of image pairs in which at least one of three raters (three board-certified pathologists) identified each morphologic feature as present in the original (dark gray) or counterfactual (light gray) image. Right, Mean pairwise inter-rater agreement (Cohen κ) per feature, computed across all pairs and both directions combined. C, Counterfactual image generation effectiveness, measured as the percentage of generated images predicted as the opposite class across varying manipulation amplitudes. D, Representative examples of ILC tile transitioning to its counterfactual IDC image and vice versa. Difference maps display pixelwise differences between the original and the synthetic tile. Scale bar applies to all images within the panel unless otherwise indicated.

Table 1 from Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology

Evaluation of diffusion autoencoders trained on different datasets.