Henry Nunoo-Mensah
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Henry Nunoo-Mensah has a total of 112 co-authors. Here are the top co-authors by number of shared publications:
- Eric Tutu Tchao
Kwame Nkrumah University of Science and Technology
11 shared publications - Andrew Selasi Agbemenu
Department of Computer Engineering Kwame Nkrumah University of Science and Technology Kumasi Ghana
9 shared publications James Dzisi Gadze
Department of Telecommunications Engineering, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana
6 shared publicationsJerry John Kponyo
Responsible Artificial Intelligence Lab Kwame Nkrumah University of Science and Technology Kumasi Ashanti Ghana
6 shared publications- Jerry John Kponyo
Kwame Nkrumah University of Science and Technology
6 shared publications Eric Tutu Tchao
Department of Computer Engineering Kwame Nkrumah University of Science and Technology Kumasi Ashanti Ghana
5 shared publications
Latest Publications
- Open Access
- 31 August 2026
- Published by Springer Nature in Discover Education
- Vol. 5 (1) , 816
A curated dataset for emotion-based analysis of students’ experiences in higher education
- 31 August 2026
- Published by Springer Nature in Discover Education
- Vol. 5 (1) , 816
Open AccessA curated dataset for emotion-based analysis of students’ experiences in higher education
- 12 August 2026
- Published by Springer Nature
- p. 3-20
Improving Speech Recognition for Ghanaian Accents Using Transformer-Based Accent Classification
- 12 August 2026
- Published by Springer Nature
- p. 3-20
Improving Speech Recognition for Ghanaian Accents Using Transformer-Based Accent Classification
- Open Access
- 4 August 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (8)
Attn‐DeCGAN: A Diversity‐Enhanced CycleGAN With Attention for High‐Fidelity Medical Image Translation
Unpaired image‐to‐image translation has emerged as a transformative paradigm in medical imaging, enabling unpaired image translation without the need for aligned datasets. While cycle‐consistent generative adversarial networks (CycleGANs) have shown considerable promise in this domain, they remain inherently constrained by the locality of convolutional operations, resulting in global structural inconsistencies, and by mode collapse, which restricts generative diversity. To overcome these limitations, we propose Attn‐DeCGAN, a novel attention‐augmented, diversity‐aware CycleGAN framework designed to enhance both structural fidelity and perceptual diversity in CT‐MRI translation tasks. Attn‐DeCGAN replaces conventional ResNet‐based generators with Hybrid Perception Blocks (HPBs), which synergise depthwise convolutions for spatially efficient local feature extraction with a Dual‐Pruned Self‐Attention (DPSA) mechanism that enables sparse, content‐adaptive modeling of long‐range dependencies at linear complexity. This architectural innovation facilitates the modeling of anatomically distant relationships while maintaining inference efficiency. The model is trained using a composite loss function incorporating adversarial, cycle‐consistency, identity, and VGG19‐based structural consistency losses to preserve both realism and anatomical detail. Extensive empirical evaluations demonstrate that Attn‐DeCGAN achieves superior performance across key metrics, including the lowest FID scores (60, 58), highest PSNR (27, 33), and statistically significant improvements in perceptual diversity (LPIPS, ) compared to state‐of‐the‐art baselines. Ablation studies underscore the critical role of spectral normalization in stabilizing adversarial training and enhancing attention effectiveness. Expert radiologist assessments confirmed the clinical superiority of Attn‐DeCGAN over the next best baseline, DeCGAN, with 100% real classifications and higher confidence scores in CT synthesis, and more anatomically convincing outputs in MRI translation. This has particular utility in low‐resource clinical environments where MRI is scarce, supporting synthetic MRI generation for diagnosis, radiotherapy planning, and medical image dataset augmentation. Despite increased training complexity, Attn‐DeCGAN retains efficient inference, positioning it as a technically robust and clinically deployable solution for high‐fidelity unpaired medical image translation.
- 4 August 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (8)
Open AccessAttn‐DeCGAN: A Diversity‐Enhanced CycleGAN With Attention for High‐Fidelity Medical Image Translation
Unpaired image‐to‐image translation has emerged as a transformative paradigm in medical imaging, enabling unpaired image translation without the need for aligned datasets. While cycle‐consistent generative adversarial networks (CycleGANs) have shown considerable promise in this domain, they remain inherently constrained by the locality of convolutional operations, resulting in global structural inconsistencies, and by mode collapse, which restricts generative diversity. To overcome these limitations, we propose Attn‐DeCGAN, a novel attention‐augmented, diversity‐aware CycleGAN framework designed to enhance both structural fidelity and perceptual diversity in CT‐MRI translation tasks. Attn‐DeCGAN replaces conventional ResNet‐based generators with Hybrid Perception Blocks (HPBs), which synergise depthwise convolutions for spatially efficient local feature extraction with a Dual‐Pruned Self‐Attention (DPSA) mechanism that enables sparse, content‐adaptive modeling of long‐range dependencies at linear complexity. This architectural innovation facilitates the modeling of anatomically distant relationships while maintaining inference efficiency. The model is trained using a composite loss function incorporating adversarial, cycle‐consistency, identity, and VGG19‐based structural consistency losses to preserve both realism and anatomical detail. Extensive empirical evaluations demonstrate that Attn‐DeCGAN achieves superior performance across key metrics, including the lowest FID scores (60, 58), highest PSNR (27, 33), and statistically significant improvements in perceptual diversity (LPIPS, ) compared to state‐of‐the‐art baselines. Ablation studies underscore the critical role of spectral normalization in stabilizing adversarial training and enhancing attention effectiveness. Expert radiologist assessments confirmed the clinical superiority of Attn‐DeCGAN over the next best baseline, DeCGAN, with 100% real classifications and higher confidence scores in CT synthesis, and more anatomically convincing outputs in MRI translation. This has particular utility in low‐resource clinical environments where MRI is scarce, supporting synthetic MRI generation for diagnosis, radiotherapy planning, and medical image dataset augmentation. Despite increased training complexity, Attn‐DeCGAN retains efficient inference, positioning it as a technically robust and clinically deployable solution for high‐fidelity unpaired medical image translation.
- 11 June 2025
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 28 (2) , 461-481
Enhancing speech recognition through diverse shared features accent classification
- 11 June 2025
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 28 (2) , 461-481
Enhancing speech recognition through diverse shared features accent classification
- Open Access
- 1 June 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (6)
Diversity in Stable GANs: A Systematic Review of Mode Collapse Mitigation Strategies
Mode collapse poses a critical challenge in training generative adversarial networks (GANs), particularly in applications such as medical imaging, where diverse and clinically relevant outputs are essential. This systematic review methodically examines the causes and impacts of mode collapse, classifies mitigation strategies into four categories; architectural modifications, loss function adaptations, regularization techniques, and hybrid techniques, and evaluates their effectiveness. Hybrid approaches, combining adversarial loss adaptation, architectural modifications, and regularization terms, are particularly promising. Additionally, integrating GANs with frameworks such as federated learning, diffusion models, and attention mechanisms shows potential to improve stability and diversity. By highlighting successful strategies and identifying gaps, especially in domain‐specific contexts such as medical imaging, this review aims to advance GAN applications in low‐resource regions and beyond, improving healthcare and other critical sectors.
- 1 June 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (6)
Open AccessDiversity in Stable GANs: A Systematic Review of Mode Collapse Mitigation Strategies
Mode collapse poses a critical challenge in training generative adversarial networks (GANs), particularly in applications such as medical imaging, where diverse and clinically relevant outputs are essential. This systematic review methodically examines the causes and impacts of mode collapse, classifies mitigation strategies into four categories; architectural modifications, loss function adaptations, regularization techniques, and hybrid techniques, and evaluates their effectiveness. Hybrid approaches, combining adversarial loss adaptation, architectural modifications, and regularization terms, are particularly promising. Additionally, integrating GANs with frameworks such as federated learning, diffusion models, and attention mechanisms shows potential to improve stability and diversity. By highlighting successful strategies and identifying gaps, especially in domain‐specific contexts such as medical imaging, this review aims to advance GAN applications in low‐resource regions and beyond, improving healthcare and other critical sectors.
- 29 May 2025
- Published by Wiley in Applied Research
- Vol. 4 (3)
Wavelet‐Enhanced Deformable Convolutional Network for Breast Cancer Classification in High‐Resolution Histopathology Images
The limitations of deep learning methods in processing high‐resolution inputs can impact the accuracy and efficiency of their results. This study presents a new architectural framework that combines wavelet‐based preprocessing with deformable convolutional networks to classify high‐resolution histopathological images. Our methodology utilizes multi‐resolution wavelet decomposition for efficient feature extraction which maintains diagnostically significant information. This improvement is augmented by deformable convolutions, which improve robustness against geometric transformations of the inputs. Empirical evaluation on the BreaKHis data set shows an image‐level accuracy of 96.47% and a patient‐level accuracy of 96.55% at 200× magnification. The architecture consistently performs well across different magnification levels, with particular efficiency at higher resolutions where detailed morphological features are essential for accurate diagnosis. Ablation studies support our key architectural contributions, including reduced computational complexity through wavelet‐based feature extraction, improved geometric invariance via deformable convolutions, and better classification performance than conventional methods. These findings suggest significant potential for improving diagnostic workflows in clinical settings where pathological expertise may be limited.
- 29 May 2025
- Published by Wiley in Applied Research
- Vol. 4 (3)
Wavelet‐Enhanced Deformable Convolutional Network for Breast Cancer Classification in High‐Resolution Histopathology Images
The limitations of deep learning methods in processing high‐resolution inputs can impact the accuracy and efficiency of their results. This study presents a new architectural framework that combines wavelet‐based preprocessing with deformable convolutional networks to classify high‐resolution histopathological images. Our methodology utilizes multi‐resolution wavelet decomposition for efficient feature extraction which maintains diagnostically significant information. This improvement is augmented by deformable convolutions, which improve robustness against geometric transformations of the inputs. Empirical evaluation on the BreaKHis data set shows an image‐level accuracy of 96.47% and a patient‐level accuracy of 96.55% at 200× magnification. The architecture consistently performs well across different magnification levels, with particular efficiency at higher resolutions where detailed morphological features are essential for accurate diagnosis. Ablation studies support our key architectural contributions, including reduced computational complexity through wavelet‐based feature extraction, improved geometric invariance via deformable convolutions, and better classification performance than conventional methods. These findings suggest significant potential for improving diagnostic workflows in clinical settings where pathological expertise may be limited.
- Open Access
- 20 March 2025
- Published by Elsevier in Intelligent Systems with Applications
- Vol. 26
Deep learning for efficient high-resolution image processing: A systematic review
- 20 March 2025
- Published by Elsevier in Intelligent Systems with Applications
- Vol. 26
Open AccessDeep learning for efficient high-resolution image processing: A systematic review
- 17 March 2025
- Published by Springer Nature in Nature Human Behaviour
- Vol. 9 (5) , 842-845
We need Afrocentric datasets for inclusive AI growth
- 17 March 2025
- Published by Springer Nature in Nature Human Behaviour
- Vol. 9 (5) , 842-845
We need Afrocentric datasets for inclusive AI growth
- Open Access
- 18 February 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (2)
Wavelet‐Based Feature Extraction for Efficient High‐Resolution Image Classification
Convolutional neural networks (CNNs) typically compress high‐resolution images to minimize computational requirements. However, this can lead to loss of information and reduced accuracy in classification tasks. This paper introduces WaveNet, a novel approach for processing high‐resolution images using wavelet‐domain inputs in CNNs. We address the challenge of maintaining classification accuracy with high‐resolution inputs while minimizing computational complexity. Our method employs wavelet packet transform (WPT) for image pre‐processing, extracting detailed multi‐scale and directional information from high‐resolution images. We propose a wavelet‐adaptive efficient channel attention (WAECA) module to dynamically select the most informative wavelet subbands. Popular CNN architectures like ResNet‐50 and MobileNetV2 are adapted by replacing their initial convolutional layers with wavelet‐transformed inputs, enabling direct learning in the wavelet domain. Experiments conducted on the Caltech‐256 and ALOT datasets demonstrate that WaveNet improves classification accuracy while reducing computational complexity. For instance, our wavelet‐enhanced ResNet‐50 achieves a Top‐1 accuracy of 72.47% on Caltech‐256, outperforming the baseline (70.65%) while reducing FLOPs from 16.52G to 3.98G. Similar improvements are observed across different architectures and datasets. We also evaluate various wavelet filters and ResNet backbones, finding that the bior1.1 filter and ResNet‐50 provide optimal performance. This work presents a practical solution for developing more accurate and efficient models for high‐resolution inputs without extensive computational resources or complex architectural modifications.
- 18 February 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (2)
Open AccessWavelet‐Based Feature Extraction for Efficient High‐Resolution Image Classification
Convolutional neural networks (CNNs) typically compress high‐resolution images to minimize computational requirements. However, this can lead to loss of information and reduced accuracy in classification tasks. This paper introduces WaveNet, a novel approach for processing high‐resolution images using wavelet‐domain inputs in CNNs. We address the challenge of maintaining classification accuracy with high‐resolution inputs while minimizing computational complexity. Our method employs wavelet packet transform (WPT) for image pre‐processing, extracting detailed multi‐scale and directional information from high‐resolution images. We propose a wavelet‐adaptive efficient channel attention (WAECA) module to dynamically select the most informative wavelet subbands. Popular CNN architectures like ResNet‐50 and MobileNetV2 are adapted by replacing their initial convolutional layers with wavelet‐transformed inputs, enabling direct learning in the wavelet domain. Experiments conducted on the Caltech‐256 and ALOT datasets demonstrate that WaveNet improves classification accuracy while reducing computational complexity. For instance, our wavelet‐enhanced ResNet‐50 achieves a Top‐1 accuracy of 72.47% on Caltech‐256, outperforming the baseline (70.65%) while reducing FLOPs from 16.52G to 3.98G. Similar improvements are observed across different architectures and datasets. We also evaluate various wavelet filters and ResNet backbones, finding that the bior1.1 filter and ResNet‐50 provide optimal performance. This work presents a practical solution for developing more accurate and efficient models for high‐resolution inputs without extensive computational resources or complex architectural modifications.
- Open Access
- 26 January 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (1)
Multi‐Wound Classification: Exploring Image Enhancement and Deep Learning Techniques
Wounds contribute to 30%–42% of hospital visits and 9% of deaths but remain underreported in Africa. Diseases and surgeries increase wound prevalence, especially in rural areas where 27%–82% of people live, and health facilities are poor or non‐existent. This research aims to design a disease‐related wound classification model for online diagnosis and telemedicine support for traditional health practitioners and village health workers. This paper focuses on wounds from diabetic ulcers, pressure ulcers, surgery, and venous ulcers. The approaches used included Contrast Limited Adaptive Histogram Equalization (CLAHE) with machine and deep learning models, Discrete Wavelet Transformations (DWT) with a novel Gated Wavelet Convolutional Neural Network (CNN) model, and FixCaps, an improved version of Capsule Networks utilizing Convolutional Block Attention Module (CBAM) to reduce spatial information loss. The performance metrics showed similar results for the first two approaches, but FixCaps was the most proficient, with accuracy, precision, recall, and F‐score of 93.83%, 95.41%, 88.63%, and 90.93% respectively. FixCaps had trainable parameters of about 8.28 MB compared with the 195.64 MB of the Gated Wavelet CNN Model.
- 26 January 2025
- Published by Wiley in Engineering Reports
- Vol. 7 (1)
Open AccessMulti‐Wound Classification: Exploring Image Enhancement and Deep Learning Techniques
Wounds contribute to 30%–42% of hospital visits and 9% of deaths but remain underreported in Africa. Diseases and surgeries increase wound prevalence, especially in rural areas where 27%–82% of people live, and health facilities are poor or non‐existent. This research aims to design a disease‐related wound classification model for online diagnosis and telemedicine support for traditional health practitioners and village health workers. This paper focuses on wounds from diabetic ulcers, pressure ulcers, surgery, and venous ulcers. The approaches used included Contrast Limited Adaptive Histogram Equalization (CLAHE) with machine and deep learning models, Discrete Wavelet Transformations (DWT) with a novel Gated Wavelet Convolutional Neural Network (CNN) model, and FixCaps, an improved version of Capsule Networks utilizing Convolutional Block Attention Module (CBAM) to reduce spatial information loss. The performance metrics showed similar results for the first two approaches, but FixCaps was the most proficient, with accuracy, precision, recall, and F‐score of 93.83%, 95.41%, 88.63%, and 90.93% respectively. FixCaps had trainable parameters of about 8.28 MB compared with the 195.64 MB of the Gated Wavelet CNN Model.