Tuesday, November 17, 2026
Science
No Result
View All Result
  • Login
  • HOME
  • SCIENCE NEWS
  • CONTACT US
  • HOME
  • SCIENCE NEWS
  • CONTACT US
No Result
View All Result
Scienmag
No Result
View All Result
Home Science News

Lightweight Attention Model Optimizes Monkeypox Detection for Resource-Constrained Clinics

November 17, 2026
in Science News
Ophelia Keating
By Ophelia Keating Scienmag Editorial Profile - Health Services Research
Reading Time: 6 mins read
0
Lightweight Attention Model Optimizes Monkeypox Detection for Resource-Constrained Clinics

Lightweight Attention Model Optimizes Monkeypox Detection for Resource-Constrained Clinics

65
SHARES
587
VIEWS
Share on FacebookShare on Twitter
ADVERTISEMENT

A research team led by Vandana and Chetna Sharma has introduced a lightweight convolutional neural network architecture optimized through genetic algorithms to accurately identify monkeypox and distinguish it from other dermatological conditions. Evaluated across 847 clinical images representing six distinct diagnostic categories, the proposed evolutionary framework—designated GA-CNN-Attn—achieved an overall classification accuracy of 98.18 percent, alongside a precision of 98.86 percent, a recall of 98.41 percent, an F1-score of 98.53 percent, and a macro-averaged receiver operating characteristic area under the curve of 99.3 percent. Published in the Journal of Transformative Technologies and Sustainable Development, the study demonstrates that coupling automated hyperparameter discovery with targeted attention modules allows compact neural architectures to rival or outperform substantially heavier deep learning models. By reducing computational demands while maintaining high diagnostic precision, the investigators aimed to address the critical shortage of accessible, automated diagnostic tools in community clinics and low-resource healthcare environments where specialized dermatologists and high-performance computing clusters remain unavailable.

The ongoing public health importance of monkeypox, officially designated as Mpox, has intensified following global outbreaks that have accumulated more than 100,000 laboratory-confirmed cases across multiple geographic regions according to World Health Organization reports cited by the authors. Caused by the monkeypox virus, the infection presents with cutaneous manifestations that closely mimic several other eruptive poxviruses and common dermatological conditions, most notably chickenpox, smallpox, and measles. In early clinical stages, superficial pustules, vesicles, and rash patterns make reliable visual differentiation exceptionally difficult for general practitioners without specialized dermatological training. This diagnostic ambiguity is magnified in remote and under-resourced medical facilities where confirmatory molecular diagnostic testing, such as polymerase chain reaction assays, is either prohibitively expensive, logistically inaccessible, or subject to prolonged processing delays. Delays in accurate identification elevate secondary transmission risks within communities, underscoring an acute global requirement for rapid, automated, and inexpensive triage systems capable of assisting healthcare workers during initial patient evaluations.

Although deep learning architectures have demonstrated remarkable capabilities in dermatological image analysis, existing diagnostic solutions frequently remain impractical for widespread edge-device implementation. State-of-the-art vision models such as VGG16, ResNet, and EfficientNet have achieved diagnostic accuracies ranging between 90 and 97 percent across various medical imaging benchmarks, but their deep topological depth demands substantial computational power, large graphics processing units, and extensive memory bandwidth. Furthermore, many conventional implementations rely on manual parameter tuning or heuristic trial-and-error procedures that fail to explore complex non-convex hyperparameter search spaces effectively. Previous models also frequently lack explicit spatial and channel attention mechanisms tailored to dermatological lesions, leaving them vulnerable to background visual artifacts such as healthy skin folds, ambient examination room lighting variations, or diverse anatomical positioning. Consequently, these resource-heavy networks exhibit degraded generalization and excessive inference latency when ported to portable screening hardware or mobile clinical platforms operating in frontline outbreak settings.

To train and rigorously benchmark their approach, the researchers utilized the publicly available Kaggle Monkeypox 2022 Remastered dataset, which comprises 847 clinical skin disease photographs partitioned across six distinct diagnostic classes. Specifically, the distribution includes 358 smallpox images, 178 chickenpox images, 160 monkeypox images, 54 cowpox images, 50 healthy skin images, and 47 measles images, reflecting substantial class imbalance. In addition to infectious conditions, the dataset incorporated benign pigmentary irregularities such as melasma to assess the network’s capacity to discriminate pathogenic eruptive lesions from non-infectious cutaneous abnormalities. All photographic inputs were standardized by resizing them to uniform dimensions of 224 by 224 pixels with three color channels and normalizing pixel intensity values to a zero-to-one range. To counteract the modest sample size and avert overfitting, the authors applied on-the-fly data augmentation during training, executing horizontal flips, vertical flips, random rotations within 20 degrees, random scaling between 0.9 and 1.1, translation shifts of up to 10 percent, and brightness adjustments.

The underlying convolutional neural network was engineered specifically to limit parameter count and forward-pass operations while preserving fine spatial feature representations. Beginning with an input layer receiving normalized three-channel images, the architecture incorporates three sequential convolutional blocks featuring receptive field kernels ranging from 3 by 3 to 5 by 5 pixels, each activated by Rectified Linear Unit functions. Batch normalization layers follow each convolution operation to stabilize training dynamics and reduce internal covariate shift, while max-pooling layers downsample spatial dimensions to restrain computational load. Following the second convolutional block, an integrated attention module dynamically weights feature activations before representations are flattened into a fully connected dense layer containing 256 neurons. An output layer governed by a softmax activation function computes probabilistic class assignments across the six categories. Operating with roughly 1.3 million trainable parameters and demanding about 210 million floating-point operations per forward pass, the model drastically cuts resource usage compared to traditional architectures.

A defining characteristic of the investigative framework is the comparative exploration and integration of lesion-focused attention modules, designed to emphasize clinically significant regions while suppressing irrelevant visual backgrounds. The authors systematically evaluated three distinct attention paradigms: Squeeze-and-Excitation blocks, which apply global average pooling to recalibrate channel dependencies; the Convolutional Block Attention Module, known as CBAM, which sequentially couples channel-wise and spatial attention operations; and transformer-style self-attention mechanisms that model long-range spatial correlations across feature maps. In the CBAM formulation, average-pooling and max-pooling operations extract spatial context, which a multi-layer perceptron processes to highlight informative feature channels, followed by a spatial convolution that maps the focal coordinates of lesions. Through empirical optimization, the hybrid CBAM structure proved most effective, as its joint channel and spatial refinement allowed the network to isolate irregular lesion boundaries and chromatic variations characteristic of monkeypox pustules far more reliably than channel-only Squeeze-and-Excitation or computationally heavy self-attention modules.

Rather than depending on arbitrary manual adjustments or resource-exhaustive grid searches, the research team employed a Genetic Algorithm to discover optimal architectural parameters, training hyperparameters, and attention configurations concurrently. Drawing upon evolutionary principles of natural selection, chromosomal strings encoded mixed discrete and continuous variables, including convolutional filter counts, dense unit counts, kernel dimensions, learning rates, batch sizes, dropout rates, optimizer types, and attention block variants. Operating across seven generations with an initial population of 10 candidate individuals, the evolutionary process utilized tournament or roulette selection, uniform crossover with a probability of 0.8, and random mutation at a 0.1 rate, alongside an elitism mechanism that preserved top-performing models across generations. Model evaluation relied on a fitness function balancing validation accuracy, macro-averaged F1-score, and categorical cross-entropy loss through defined weighting coefficients. The resulting evolutionary search identified an optimal configuration featuring the CBAM attention module, the Adam optimizer, a learning rate of 0.001, a batch size of 40, a 3 by 3 kernel size, and a dropout rate of 0.3.

Experimental validation was executed on a workstation equipped with an Intel Core i7 processor, 16 gigabytes of random-access memory, and an NVIDIA graphics processing unit running TensorFlow and Keras. To ensure diagnostic reliability despite marked class skew, the investigators employed a stratified five-fold cross-validation scheme alongside an initial dataset split of 70 percent training, 15 percent validation, and 15 percent held-out testing, incorporating class-weighted cross-entropy loss and early stopping with a 10-epoch patience threshold. The resulting performance proved remarkably consistent, demonstrating an average accuracy of 98.18 percent, precision of 98.86 percent, recall of 98.41 percent, and an F1-score of 98.53 percent across cross-validation folds. In receiver operating characteristic evaluations, every individual cutaneous class achieved an area under the curve exceeding 98 percent, culminating in a macro-averaged score of 99.3 percent. Close alignment between training and validation curves, with an accuracy divergence under 1.5 percent, confirmed that regularization safeguards prevented noticeable overfitting.

When benchmarked against established convolutional and transfer-learning architectures evaluated under identical data-split and augmentation protocols, the GA-CNN-Attn framework demonstrated clear diagnostic superiority. Baseline networks exhibited notable vulnerabilities: standard ResNet and VGG16 models registered accuracies of only 51 percent and 45 percent respectively, suffering from parameter over-parameterization and convergence difficulties on the compact dataset. Intermediate architectures such as Xception reached 91 percent accuracy, while lightweight competitors MobileNetV2 and EfficientNet-B0 attained 94.5 percent and 97.0 percent accuracy. Non-parametric Wilcoxon signed-rank tests across cross-validation folds confirmed that the performance advantages of GA-CNN-Attn were statistically significant at alpha equals 0.05, rejecting the null hypothesis of random variation. Furthermore, compared to EfficientNet-B0, the proposed model delivered a 1.2 percentage point improvement in accuracy while maintaining roughly three times fewer parameters and requiring approximately 1.9 times fewer floating-point operations, translating into markedly reduced CPU and GPU latency during inference.

From a translational perspective, the compact parameter footprint and rapid inference speed position the GA-CNN-Attn model as a practical clinical decision-support tool capable of deployment on standard computers, low-cost hardware, and mobile platforms. Frontline triage personnel equipped with basic smartphone cameras could capture preliminary skin lesion photographs to screen suspected viral infections and expedite referral prioritization for dermatological consultation during regional epidemics. Nevertheless, the investigators delineated several key methodological constraints that require addressing prior to clinical adoption. The experimental findings remain anchored to a single benchmark repository consisting of 847 images, necessitating future validation across multi-center, geographically diverse cohorts with broader variations in skin phototypes and capture conditions. Additionally, the current model functions largely as a black-box system; subsequent research must integrate explainable artificial intelligence frameworks, such as Grad-CAM saliency maps or SHAP attribution methods, alongside prospective clinical trials to establish interpretability, clinician trust, and real-world operational safety.

Subject of Research: Evolutionary optimization of lightweight attention convolutional neural networks for multi-class skin lesion and monkeypox diagnosis

Article Title: A Lightweight Attention CNN Model for Monkeypox Detection with Evolutionary Optimization

Article References: Vandana, & Sharma, C. (2026). A Lightweight Attention CNN Model for Monkeypox Detection with Evolutionary Optimization. Journal of Transformative Technologies and Sustainable Development, 10(1), Article 3. https://doi.org/10.1007/s41314-026-00092-x

Image Credits: AI Generated

DOI: 10.1007/s41314-026-00092-x

Keywords: Monkeypox detection, Mpox diagnosis, Convolutional neural networks, Genetic algorithms, Attention mechanisms, CBAM, Lightweight deep learning, Skin disease classification, Medical image processing, Hyperparameter optimization, Evolutionary computing, Resource-constrained healthcare, Dermatological imaging, Cross-validation, Class imbalance, Diagnostic decision support, Wilcoxon signed-rank test

Cite Scienmag News

Ophelia Keating. (November 17, 2026). Lightweight Attention Model Optimizes Monkeypox Detection for Resource-Constrained Clinics. Scienmag. https://scienmag.com/lightweight-attention-model-optimizes-monkeypox-detection-for-resource-constrained-clinics/

Ophelia Keating. "Lightweight Attention Model Optimizes Monkeypox Detection for Resource-Constrained Clinics." Scienmag, 17 November 2026, https://scienmag.com/lightweight-attention-model-optimizes-monkeypox-detection-for-resource-constrained-clinics/. Accessed 17 November 2026.

Ophelia Keating. "Lightweight Attention Model Optimizes Monkeypox Detection for Resource-Constrained Clinics." Scienmag. November 17, 2026. https://scienmag.com/lightweight-attention-model-optimizes-monkeypox-detection-for-resource-constrained-clinics/

Tags: attention mechanismsCBAMclass imbalanceconvolutional neural networkscross-validationDermatological imagingDiagnostic decision supportEvolutionary computinggenetic algorithmshyperparameter optimizationlightweight deep learningmedical image processingMonkeypox detectionMpox diagnosisresource-constrained healthcareSkin disease classificationWilcoxon signed-rank test
Share26Tweet16
Previous Post

Single Programmable Photonic Processor Executes Column-Streaming Optical Convolution With High Accuracy

Next Post

Russian Regional Decarbonization Study Shows Severe Emission and Vulnerability Disparities

Related Posts

Rapid Prenatal Whole-Genome Sequencing Improves Genetic Diagnosis Turnaround in Multicenter Cohort
Science News

Rapid Prenatal Whole-Genome Sequencing Improves Genetic Diagnosis Turnaround in Multicenter Cohort

November 17, 2026
Realist Synthesis Uncovers Core Mechanisms Enabling External Facilitation in Healthcare Interventions
Science News

Realist Synthesis Uncovers Core Mechanisms Enabling External Facilitation in Healthcare Interventions

November 17, 2026
Lie Group Analysis Uncovers Exact Solutions for the Nonlinear VK System
Science News

Lie Group Analysis Uncovers Exact Solutions for the Nonlinear VK System

November 17, 2026
Early Tildrakizumab Treatment Predicts Complete Psoriasis Clearance in Five-Year Italian Study
Medicine

Early Tildrakizumab Treatment Predicts Complete Psoriasis Clearance in Five-Year Italian Study

November 17, 2026
Obsessive-Compulsive Disorder Patients Show Altered Emotional Processing and Oculomotor Control Patterns
Medicine

Obsessive-Compulsive Disorder Patients Show Altered Emotional Processing and Oculomotor Control Patterns

November 17, 2026
Four Key Clinical Factors Associated With Heart Injury in Gastrointestinal Bleeding
Medicine

Four Key Clinical Factors Associated With Heart Injury in Gastrointestinal Bleeding

November 17, 2026
Next Post
Russian Regional Decarbonization Study Shows Severe Emission and Vulnerability Disparities

Russian Regional Decarbonization Study Shows Severe Emission and Vulnerability Disparities

  • Mothers who receive childcare support from maternal grandparents show more optimized

    Mothers who receive childcare support from maternal grandparents show more parental warmth, finds NTU Singapore study

    27656 shares
    Share 11059 Tweet 6912
  • University of Seville Breaks 120-Year-Old Mystery, Revises a Key Einstein Concept

    1061 shares
    Share 424 Tweet 265
  • Bee body mass, pathogens and local climate influence heat tolerance

    682 shares
    Share 273 Tweet 171
  • Researchers record first-ever images and data of a shark experiencing a boat strike

    546 shares
    Share 218 Tweet 137
  • Groundbreaking Clinical Trial Reveals Lubiprostone Enhances Kidney Function

    531 shares
    Share 212 Tweet 133
Science

Embark on a thrilling journey of discovery with Scienmag.com—your ultimate source for cutting-edge breakthroughs. Immerse yourself in a world where curiosity knows no limits and tomorrow’s possibilities become today’s reality!

RECENT NEWS

  • Rapid Prenatal Whole-Genome Sequencing Improves Genetic Diagnosis Turnaround in Multicenter Cohort
  • Realist Synthesis Uncovers Core Mechanisms Enabling External Facilitation in Healthcare Interventions
  • Lie Group Analysis Uncovers Exact Solutions for the Nonlinear VK System
  • Early Tildrakizumab Treatment Predicts Complete Psoriasis Clearance in Five-Year Italian Study

Categories

  • Agriculture
  • Anthropology
  • Archaeology
  • Athmospheric
  • Biology
  • Biotechnology
  • Blog
  • Bussines
  • Cancer
  • Chemistry
  • Climate
  • Earth Science
  • Editorial Policy
  • Marine
  • Mathematics
  • Medicine
  • Pediatry
  • Policy
  • Psychology & Psychiatry
  • Science Education
  • Science News
  • Social Science
  • Space
  • Technology and Engineering

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 5,150 other subscribers

© 2026 Scienmag - Science Magazine

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • HOME
  • SCIENCE NEWS
  • CONTACT US

© 2026 Scienmag - Science Magazine

Discover more from Science

Subscribe now to keep reading and get access to the full archive.

Continue reading