Controllable image structure editing has attracted increasing attention. While recent interactive point-based methods are convenient and realistic, they often lack fine-grained control over localized content. Partial sketches provide a simple yet expressive interface for local structure manipulation. However, existing partial-sketch-based manipulation methods relying on generative adversarial networks (GANs) suffer from limited generalization and fidelity. Moreover, although diffusion-based adapters excel at global conditioning (e.g., edge maps), localized editing with partial strokes remains challenging due to two key issues: effectively injecting sparse stroke conditions during denoising and preserving non-edited regions to avoid unintended changes. To address these challenges, we propose DiffStroke, a mask-free framework for localized image manipulation with partial sketches. We introduce trainable Image-Stroke Fusion (ISF) blocks to fuse source images and strokes at the feature level, enabling precise local shape control while maintaining appearance consistency. We further develop a self-supervised mask estimator to protect irrelevant regions without manual input. Specifically, we leverage Tweedie's formula to estimate a clean latent image from noisy latents, blend the denoised result with the source, and train the mask estimator by minimizing the error between the blended latent and the target latent. Experiments on natural and facial images demonstrate that DiffStroke outperforms state-of-the-art methods on both simple and complex stroke-based editing tasks. DiffStroke can also be combined with text prompts to produce diverse and creative results. Code is available at https://github.com/CMACH508/DiffStroke.
Historical sketch authentication is challenging because securely attributed reference sets are often small, and stylistic evidence is carried primarily by line, texture, tonal variation, and mark-making. We present a reproducible framework for verifying historical sketches using artist-specific one-class autoencoders trained on compact handcrafted feature representations. Ten artist models were trained using authenticated sketches from six open-access cultural heritage collections. Each drawing was represented by five interpretable descriptors, namely, Fourier-domain energy, Shannon entropy, global contrast, Grey-Level Co-occurrence Matrix homogeneity, and box-counting fractal complexity. The system was evaluated using a biometric-style verification protocol in which each artist model was tested on genuine held-out works and impostor works by other artists. On the primary evaluation partition of 900 decisions, comprising 90 genuine and 810 impostor trials, the method achieved 87.6% balanced accuracy, 77.8% True Acceptance Rate, 2.6% False Acceptance Rate, 0.748 Matthews Correlation Coefficient, and 11.4% Equal Error Rate. Performance remained stable across 20 repeated random train/test splits. The proposed model also outperformed Gaussian and one-class SVM baselines, while pretrained ResNet50 and EfficientNet-V2 feature representations performed substantially worse in this data-scarce setting. Leave-one-feature-out ablation confirmed that all five descriptors contributed positively, with fractal complexity and GLCM homogeneity providing the strongest individual contributions. Error analysis revealed structured false-accept pathways to be consistent with stylistic proximity between artists. The framework provides transparent, reproducible, and interpretable quantitative evidence for historical sketch verification. It is intended to support, not replace, expert connoisseurship in attribution settings where available reference corpora are limited.
Sketches are an efficient and effective tool for generating 3D human meshes with arbitrary body shapes and poses. However, current mesh reconstruction methods are mainly designed for natural images, which are hard to apply to sketches due to the abstract and sparse characteristics of the latter. Moreover, there is no dataset with sufficient sketch-mesh pairs for developing and evaluating relevant methods. To tackle these issues, we introduce a hybrid framework that fits parametric human models (e.g., skinned multi-person linear model) to sketches in a coarse-to-fine manner. Specifically, the proposed framework consists of three core components: (i) Given a sketch image as the input, a vision transformer-based Local Image Encoder (LIE) is introduced to model the local structures of the sketch and yields a coarse mesh estimation. (ii) A Global Point Encoder (GPE) taking the 2D coordinates of sketch contours as inputs, is also utilized to obtain the global representation of the sketch. (iii) As the local presentation can depict human poses more precisely while the global representation is more suitable for body shapes, we propose a graph-based refiner (GRefiner) to leverage the advantages of both representations and generate the final well-fitted mesh. Furthermore, we collect a large-scale dubbed Sketch3DS, containing approximately 10,000 paired sketches and human meshes with diverse poses and shapes. Extensive experiments on Sketch3DS demonstrate that the proposed approach outperforms existing methods, achieving accurate alignment between input sketches and constructed human meshes.
Epidemiological studies show an increased risk for premature maternal cardiovascular disease in women after pregnancy complications, like preeclampsia, gestational hypertension and gestational diabetes mellitus. Our goal was to create a novel "digital companion" for women with such pregnancy complications, in the format of a mobile health-assisted user-centered follow-up software application (app). A cardiovascular postpartum follow-up program was developed as a digital companion, including a new mobile application (app), which is based on Norwegian obstetric and international guidelines. The MumCare app was developed through a co-creation process that included users, stakeholders, and clinical experts. Five qualitative interviews and 10 qualitative co-creative user testing interviews were conducted throughout the development stages to improve the perceived usefulness of the companion. The objective of the present study was to analyze the iterative co-creation process including users, stakeholders and clinical experts. Phase 1 involved developing the companion within an interdisciplinary expert group through an iterative process in close dialogue with users. Explorative user interviews in Phase 2 (n = 5) supported the translation of guidelines into a structured app format, visualized as MumCare sketches for design, functionality and user communication. During Phase 3, the app sketches were revised in collaboration with users, in application interviews (n = 7). During Phase 4, the programmed prototype was refined through feedback from pilot users (n = 3). The user groups highlighted several app benefits, including a follow-up system of personal modifiable risk factors, a user-friendly system for tracking blood pressure over time, with individualized feedback and prompts. The use of non-ambiguous language and symbols was appreciated among users, who also contributed new content items to the app. User-centered co-creation improved several important features during the companion development process. The MumCare app is being tested in a prospective randomized controlled clinical trial that started in June 2024. Clinicaltrials.gov reg., identifier NCT05835596.
There is incredible beauty in the microscopic organisation and structure of cells and tissues in the body. Historically, scientists seamlessly incorporated art into the reports of their findings through detailed sketches and drawings, but this mutualistic relationship is now considered unusual as science and art occupy distinct and different realms. Histology, the study of microscopic tissues, lends itself to the synthesis of art and science through colourful stains, textures, and shapes. In this paper, we revisit these past practices as we explore the benefits of blending art and science as a means of communicating microscopic anatomy, and provide examples of histological art in a variety of mediums including watercolour and graphite sketches in order to illustrate this synergistic relationship.
This study proposes a conditional hierarchical generative adversarial network (GAN) model for intelligent visual design, aiming to explore new modes of human-machine collaboration and promote digital transformation in the design field. The proposed model consists of three core modules: a multimodal conditional fusion encoder, a hierarchical generator, and a collaborative multi-discriminator architecture. It innovatively integrates multiple sources of design intent-including textual descriptions, hand-drawn sketches, spatial layouts, and brand guidelines-through a cross-modal attention mechanism. The model adopts a two-stage generation strategy of "structure first, details later" and introduces multi-objective adversarial training to simultaneously optimize generation quality, layout accuracy, and semantic consistency. Experiments on the Common Objects in Context with Stuff Annotations (COCO-Stuff) 164 K dataset show that the model achieves a Fréchet Inception Distance (FID) score of 18.37 and 92.8% layout control accuracy on the test set. These results demonstrate that the proposed model significantly outperforms mainstream baseline models in both generation quality and controllability. This study provides an innovative technical solution for professional visual design and offers theoretical and practical insights into the evolution of creative agency, human-machine collaboration mechanisms, and the pathways of digital transformation in the Artificial Intelligence era.
This study introduces a computational method for designing realistic, geometrically controlled three-dimensional (3-D) micromodels of porous media to investigate fluid flow in hydrocarbon reservoirs. The methodology utilizes a virtual framework of cubes where an arbitrary, continuous 3-D pore network is generated via two-dimensional (2-D) sketches. A key strength of this deterministic, cube-by-cube approach is the ability to independently control porosity and permeability by adjusting channel size and connectivity, facilitating the systematic study of spatial heterogeneity. Six digital models were developed with porosities ranging from 18.4% to 44.4%. Unlike traditional stochastic algorithms, this explicit geometric control enabled the accurate extraction of pore volume distributions and the establishment of a robust power-law relationship between localized porosity and specific surface area. Statistical analysis confirmed a linear correlation between porosity and pore dimensions. While focusing on design and validation, these models are 3-D printable and provide exact boundary conditions for CFD simulations. Single-phase simulations confirmed the capability to decouple absolute permeability from porosity. Consequently, this framework bridges the gap between numerical simulations and physical laboratory experiments to optimize Enhanced Oil Recovery (EOR) processes.
Top-k estimation has been a significant research focus in network data stream processing, owing to its broad range of applications. However, the rapid detection of top-k frequent flows in massive network traffic poses considerable challenges, primarily due to the stringent requirements for high-speed packet processing and constraints in resource availability. Invertible sketches, as summary data structures, enable the recovery of top-k flows with limited memory usage and bounded error guarantees. To the best of our knowledge, most existing invertible sketch algorithms suffer from high memory access overhead, which adversely affects their performance in top-k estimation. This article proposes Gentle-Sketch, a high-performance and compact invertible sketch that supports top-k estimation using small and statically allocated memory. Designed around the inherent characteristics of data streams, Gentle-Sketch adopts an adaptive structure that organizes multiple buckets in a specific manner. The entry size per bucket is tailored to match the stream distribution. Gentle-Sketch hashes each flow to multiple buckets and flexibly relocates overflowing flows, improving memory utilization. This mechanism preserves elephant flows and incorporates more mice flows without throughput loss. Extensive experimental results demonstrate that Gentle-Sketch achieves invertibility while maintaining high accuracy and throughput in top-k estimation compared to existing sketch algorithms. In particular, compared to the state-of-the-art Double-Anonymous Sketch, Gentle-Sketch improves estimation precision by over 20% and increases throughput by more than a factor of two.
Estimating mutation rates between evolutionarily related sequences is a central problem in molecular evolution. Due to the rapid expansion of datasets, modern methods avoid costly alignment and instead compare sketches of sets of constituent k-mers. While these methods perform well on many sequences, they are not robust to highly repetitive sequences such as centromeres. We present three new estimators that are robust to the presence of repeats. The estimators are applicable in different settings, depending on whether count information is available from zero, one, or both sequences. We evaluate our estimators empirically using highly repetitive alpha satellite sequences. Each estimator performs best within its class, and our strongest estimator outperforms all other tested estimators. Our software is open-source and freely available at https://github.com/medvedevgroup/Accurate_repeat-aware_kmer_based_estimator.
Scaffolding authentic assessments through a Forensic Science programme are crucial for allowing HEI students the opportunity to actively repeatedly learn and improve their knowledge and skills. This is achieved through the common sequential forensic investigative cycles of investigative review, activity plan and design, through to processing simulated forensic crime scene(s), the resulting data processing, analysis, interpretation and professional report writing investigative stages. Authentic simulated crime scene activities need to be carefully constructed to ensure case realism, use current investigative practices, grounded in pedagogic theory and scaffolded for the appropriate student learning level, albeit being conscious of the need to be balanced with potential resource/funding limitations. This article details scaffolding of deliberately different simulated forensic crime scene investigations progressively through a UK HEI forensic science programme, from 'case' intelligence and desk-based studies, through increasingly complicated indoor/outdoor crime scene data collection/processing and professional report writing investigative stages. HEI academics have direct experience of these to give assessments real authenticity. The first case details an indoor commercial bar scene, with L4 first year undergraduate students tasked with investigating and virtually recording criminal evidence, scenes and producing a virtual resource for potential presentation in court. A second case details an outdoor simulated human remains scatter scene, with L5 second year undergraduate students tasked with investigating, recording and producing scaled sketches. A third case details a wildlife forensics scene, with L6 final year undergraduate students tasked with leading the investigation, recording and producing a professional report of illegal disturbance of a badger sett. A fourth case details an outdoor mass grave investigation scene, with L7 post-graduate taught students tasked with the complete multi-staged site investigation process, from desktop study through to field reconnaissance, non-invasive data collection, through to physical excavation, forensic recovery of human remains and associated material and a professional report to be generated. All assessments received very high student feedback and provided evidence that these resources were effective for their learning and understanding in a forensic science context. These types of authentic assessments of simulated crime scenes, whilst costly in terms of development time and staff resource, will assist HEI students with crucial experience and problem-solving skills needed in time restricted scenarios to mimic those they will face in future forensic science practitioner employment. Plentiful online resources and some cost-saving suggestions for colleagues if intending to construct similar authentic assessments are also included.
This study aims to evaluate and compare the effectiveness of the Binocular Indirect Ophthalmoscope (BIO) and the Wide-Field Fundus Imaging System (RetCam) in the diagnosis of Retinopathy of Prematurity (ROP). We retrospectively reviewed routinely collected records from a standardized paired-screening workflow for preterm infants examined for ROP in outpatient and inpatient settings during 2020-2021. BIO was performed first by a senior pediatric ophthalmologist, followed within 10 min by RetCam3 imaging by another ophthalmologist. The paired BIO sketches and RetCam images were reviewed by three experienced ophthalmologists, and disagreements were resolved by consensus to establish an adjudicated reference standard for this study. The primary outcome was diagnostic accuracy for ROP zone, stage, plus/pre-plus disease, and lesion extent. A total of 796 eyes from 398 preterm infants (birth weight 1323.8 +/- 456.2 g; gestational age 29.8 +/- 2.9 weeks) were examined using both BIO and RetCam. Compared with the adjudicated reference standard, BIO and RetCam achieved AUCs of 0.997 and 0.995 for detecting any ROP, 0.983 and 0.982 for zone classification, and 0.999 and 0.959 for stage classification, respectively. RetCam misdiagnoses were most common in zone III and stage 1 disease. BIO showed lower agreement than RetCam for lesion extent (p < 0.001). BIO and RetCam showed complementary strengths in ROP screening. BIO provided clearer peripheral three-dimensional assessment of subtle stage and plus/pre-plus findings, whereas RetCam provided objective wide-field documentation of lesion distribution. A combined strategy may improve diagnostic confidence and clinical management for ROP.
K-mer-based analysis of genomic data is ubiquitous, but the presence of repetitive k-mers continues to pose problems for the accuracy of many methods. For example, the Mash tool (Ondov et al 2016) can accurately estimate the substitution rate between two low-repetitive sequences from their k-mer sketches; however, it is inaccurate on repetitive sequences such as the centromere of a human chromosome. Follow-up work by Blanca et al. (2021) has attempted to model how mutations affect k-mer sets based on strong assumptions that the sequence is non-repetitive and that mutations do not create spurious k-mer matches. However, the theoretical foundations for extending an estimator like Mash to work in the presence of repeat sequences have been lacking. In this work, we relax the non-repetitive assumption and propose a novel estimator for the mutation rate. We derive theoretical bounds on our estimator's bias. Our experiments show that it remains accurate for repetitive genomic sequences, such as the alpha satellite higher order repeats in centromeres. We demonstrate our estimator's robustness across diverse datasets and various ranges of the substitution rate and k-mer size. Finally, we show how sketching can be used to avoid dealing with large k-mer sets while retaining accuracy. Our software is available at https://github.com/medvedevgroup/Repeat-Aware_Substitution_Rate_Estimator.
A hand drawn atlas of the embryonic chick cerebellum, created by a fourth-year post-secondary science student as part of her independent research project, served the dual goal of providing an accurate and accessible map of the anatomical structure of the chick cerebellum for learning embryonic neural development, while also highlighting the benefits of art and drawing as a means to aid understanding and appreciating complex, three-dimensional neuroanatomical structures. The cerebellum is a region of the brain that is convoluted and intricately folded and it can be challenging to connect two-dimensional coronal or sagittal planes of section to the wholemount structure. This can be further complicated by anatomical variation among different species. Microscopy can be prohibitive with respect to equipment, skills and slide quality and preservation and a lack of detailed atlases of the embryonic chick brain compounds the difficulty of learning the anatomy of the pre-hatch avian cerebellum. To our knowledge, using hand-drawn graphite sketches to explore and learn about an embryonic cerebellum is a novel approach. Our initial findings suggest that using art to understand the organisation and structure of the avian cerebellum has the potential to be an effective and beneficial approach for students learning neuroanatomy.
In light of the rapidly growing large-scale data in federated ecosystems, the traditional principal component analysis (PCA) is often not applicable due to privacy protection considerations and large computational burden. Algorithms were proposed to lower the computational cost, but few can handle both high dimensionality and massive sample size under distributed settings. In this paper, we propose the FAst DIstributed (FADI) PCA method for federated data when both the dimension d and the sample size n are ultra-large, by simultaneously performing parallel computing along d and distributed computing along n . Specifically, we utilize L parallel copies of p -dimensional fast sketches to divide the computing burden along d and aggregate the results distributively along the split samples. We present a general framework applicable to multiple statistical problems, and establish comprehensive theoretical results under the general framework. We show that FADI accelerates the computation while enjoying the same non-asymptotic error rate as the traditional PCA when L p ≥ d . We also derive inferential results that characterize the asymptotic distribution of FADI, and show a phase-transition phenomenon as L p increases. We perform extensive simulations to empirically validate our theoretical findings, and apply FADI to the 1000 Genomes data to study the population structure.
Art and surgery have long shared a symbiotic relationship, each rooted in observation, precision and creativity. I was reminded of this early as a medical student during my cardiothoracic surgery and general surgery rotations, when I began making 2-min 'thumbnail' sketches after cases-just enough lines to capture planes, landmarks and instrument angles. What started as a way to stay focused quickly changed how I saw operative anatomy: messy, layered, dynamic and rarely like the textbook plate. Although drawing has anchored surgical understanding from Vesalius and Da Vinci to Cheselden and Bell, the habit has faded in modern curricula, displaced by digital imaging and largely passive viewing, especially as a medical student in surgical rotations. My experience suggests the opposite approach is both feasible and valuable. Brief, structured sketching sharpened my spatial reasoning and helped me retain operative steps; the gaps in my redraws mapped exactly to the questions I asked on rounds the next morning. Drawing not only refines visual-spatial skills but also promotes deeper cognitive processing and long-term retention-qualities essential for surgical trainees. Despite its proven benefits, many students and clinicians shy away from drawing due to perceived lack of skill, time constraints or undervaluing of artistic learning. Integrating structured drawing exercises and art-based reflection into surgical training may not only enhance anatomical comprehension but also foster mindfulness, empathy and critical observation. Reviving art within surgical education is, therefore, not an indulgence in aesthetics, but an investment in cultivating more perceptive, reflective and precise surgeons.
The archival collection of Leonard T. Furlow provides an intimate view into the mind of a surgeon whose contributions shaped modern cleft palate repair. Through a qualitative review of his teaching slides, annotated diagrams, and personal notebooks, this work provides insight into not only the surgical mind of Furlow, but also the broader patterns of creativity that defined his approach to problem solving. This study highlights his emphasis on functional outcomes, evaluation of current techniques, and proposal for new surgical methods. These efforts culminated in the development of the double opposing Z-plasty, a method that redefined surgical management of cleft palate. Beyond surgical knowledge and curiosity, his archives reveal a pattern of broad creativity, with numerous sketches and ideas to improve everyday problems. Together, these materials illustrate Furlow as a lifelong problem-solver, illustrating how curiosity and inventive thinking undermine meaningful surgical innovation.
These contrasting images are presented by Philip Alexander, MD, a native Texan, retired physician, and accomplished musician and artist. After 41 years as an internal medicine physician, Dr. Phil retired from his practice in College Station in 2016. A lifelong musician and former music professor, he often performs as an oboe soloist for the Brazos Valley Symphony Orchestra. He began exploring visual art in 1980, evolving from pencil sketches-including an official White House portrait of President Ronald Reagan-to the computer-generated drawings featured in this journal. His images, which first appeared in this journal in the spring of 2012, are his own original creations. If you would like to see your art published in the Methodist DeBakey Cardiovascular Journal, submit your creation online at journal.houstonmethodist.org as a "Humanities" entry.
AI-driven skin lesion diagnosis systems are revolutionizing dermatology practice but perform worse on darker skin populations, which threatens diagnostic equity in dermatology. Existing debiasing strategies rely on explicit skin tone annotations or adversarial removal of demographic information, which may be unavailable in practice and can harm diagnostic accuracy. We aimed to design a dermatology AI system that reduces skin-tone-related performance disparities across diverse skin populations without using skin tone labels during training. We propose a novel sketch-guided multimodal fusion framework that combines color (RGB) images with algorithmically generated structural sketches. Separate encoders extract representations from each modality, which are integrated by a gated fusion module that adaptively weighs color and structure features. A feature distillation loss aligns color features with their sketch counterparts to encourage structure-aware representations while retaining clinically relevant color cues. We trained and evaluated the model on the Fitzpatrick17k and Diverse Dermatology Images (DDI) datasets. The fairness performance was assessed with Equalized Opportunity, Equalized Odds, and Predictive Quality Disparity across skin tone groups. Out-of-domain robustness was examined using a DermaAmin and Atlas Dermatologico split. On Fitzpatrick17k, the proposed model yielded competitive accuracy and F1-score, while showing lower mean disparity in fairness evaluation than baseline methods. It also demonstrated reduced subgroup disparity across skin tone groups on the evaluated fairness metrics. In the out-of-domain evaluation setup, the model also exhibited improved fairness. On the DDI dataset, the framework showed consistent performance across different skin tone groups with reduced variation. Our proposed model shows promise in reducing skin-tone-related bias in dermatology while preserving utility without relying on explicit skin tone annotations. The observed improvements in skin tone fairness across two datasets suggest that our approach may help reduce measured subgroup disparities in automated skin lesion assessment, although clinical utility and real-world impact remain to be established through prospective validation.
Many practical security applications rely on inherently noisy data sources, such as biometric measurements, sensor-derived features, and physical unclonable functions (PUFs). This variability makes it difficult to develop reliable cryptographic primitives. Conventional solutions-like fuzzy commitment schemes, fuzzy vaults, secure sketches, and fuzzy extractors-enable key binding or generation from noisy data by combining cryptography with error-correcting codes to handle variations. However, using ECCs can cause information leakage. In this paper, we construct ISNR-PQC, an isometry noise resilient post-quantum cryptographic primitive designed for natural noisy sources, using secure biometric data as an application. ISNR-PQC specifically uses lattice-based cryptography based on the National Institute of Standards and Technology (NIST) standard FIPS 203. Our approach provides security guarantees against quantum adversaries under the Learning with Errors (LWE) assumption. We formalise correctness, robustness, and [Formula: see text]-restricted IND-CPA indistinguishability in a unified security model. Our ISNR-PQC is noise resilient because it accepts ancillary data that is close to the original encrypting data, as measured by an isometry-based threshold mapping. Our results demonstrate that ISNR-PQC addresses a major challenge in biometric authentication theory and is a promising base for future biometric and noisy-data security systems.
Rational constructivism is a contemporary theory of cognitive development. It proposes that children generate, tweak, and radically revise theories, beliefs, and representations by integrating evidence with existing knowledge via probabilistic inferential learning mechanisms. We begin with a general overview of the theory of rational constructivism, covering its theoretical commitments and predictions. We then describe how it accounts for the empirically observed patterns of both incremental and radical developmental change and discuss the cognitive mechanisms underlying those conceptual changes. Finally, we sketch out a general computational framework for it and address open questions and directions for future research.