Breast cancer (BC) remains a leading cause of cancer-related mortality globally, especially due to its metastatic tendencies. Metastatic breast cancer (MBC) occurs when cancer cells spread from the breast to other parts of the body, complicating treatment and reducing survival rates. Predicting MBC is crucial for timely intervention, but challenges persist due to noisy features that limit model accuracy. While traditional machine learning techniques have been applied to predict MBC, they often struggle with identifying key factors for accurate prediction. In this study, we propose enhancing the JAYA algorithm by incorporating a sinusoidal chaotic map for population initialization and to improve exploration during optimization. Specifically, the sinusoidal chaotic initialization is employed to generate a more uniformly distributed and diverse initial population, thereby improving search space coverage, reducing premature convergence, and enhancing the algorithm’s ability to identify relevant features for MBC prediction. The enhanced JAYA algorithm was combined with an artificial neural network (ANN) for prediction, and its performance was evaluated using a 5-year MBC dataset obtained from the Kaggle repository, comprising 6,726 instances, 26 features, and a binary class label. The results indicate that the proposed method achieves improved performance across key evaluation metrics, including accuracy, F1 score, sensitivity, and specificity, when compared with the baseline study of Muhammed et al., (2024). Specifically, the proposed method attained an accuracy of 79.76%, an F1 score of 80.13%, a sensitivity of 81.88%, and a specificity of 77.52%, compared to 79.28%, 79.66%, 81.04%, and 77.51%, respectively, reported in the baseline study, thereby demonstrating consistent performance improvements across all evaluation metrics. These improvements were confirmed to be statistically significant using a T-test. Hyperparameter tuning, including adjustments to population size and iteration count, further optimized the method’s performance, confirming the benefits of fine-tuning in metaheuristic algorithms for MBC prediction.

Deep learning has driven remarkable advances in the computer vision, but achieving state-of-the-art performance increasingly usually depends on a diverse set of techniques applied across the entire model development lifecycle. The existing surveys typically focus on isolated aspects such as data augmentation, optimization strategies, or model compression and ensemble methods. A unified understanding of how performance-boosting techniques interact across pre-training, training and post-training stages remains limited. This paper presents a comprehensive, lifecycle-oriented review of techniques aimed at improving the accuracy, generalization, robustness, and efficiency of deep learning models for computer vision. We systematically organize performance-enhancement methods into three stages, including pre-training, training and post-training, and also analyse how choices made at each stage influence downstream optimization, deployment and real-world performance. Beyond cataloguing existing methods, the review synthesizes cross-stage dependencies, trade-offs and complementary effects that are often overlooked in stage-specific surveys. In addition, we identify open challenges related to data bias, reproducibility, scalability, robustness under distribution shift and sustainability, and outline emerging research directions including automated performance-boosting pipelines, eco-efficient training, multimodal foundation models, continual learning and human-in-the-loop optimization. By providing a holistic and integrative perspective, this review aims to serve as a reference framework for researchers and practitioners seeking to design high-performance, reliable and deployable computer vision systems.

Tomatoes are one of the most widely grown horticultural crops but are highly susceptible to leaf diseases that could significantly affect yield and quality. Early and accurate disease detection is necessary to enhance crop productivity and food security. The present study presents a deep learning system for automatic tomato leaf disease classification using a weighted ensemble of EfficientNetV2 models. Seven variants of EfficientNetV2 (B0-B3, S, M, and L) were fine-tuned on an augmented version of the filtered PlantVillage dataset. To improve feature visibility under varying lighting conditions, images were preprocessed using contrast-limited adaptive histogram equalization. Five optimization algorithms, which include Adam, Adamax, AdamW, Nadam, and RMSProp, were tested to assess their impact on model convergence and generalization. Among individual models, Adam-optimized EfficientNetV2L performed the best in accuracy with a measure of 99.50%. For classifiability resilience improvement, 16 ensemble configurations (eight unweighted and eight weighted ensembles using different combinations of EfficientNetV2-S, M, and L variants and five optimizers) were explored, with the best weights discovered using brute-force grid search. The best-performing weighted ensemble achieved 99.89% in accuracy, precision, recall, and F1-score, which was evaluated on a held-out test set not used during training or validation, demonstrating its strong generalizability. The proposed framework offers a scalable and reliable solution for early tomato disease detection with significant potential for real-time plant health monitoring in precision agriculture.Received: 10 September 2025 | Revised: 22 December 2025 | Accepted: 6 February 2026Conflicts of InterestThe authors declare that they have no conflicts of interest to this work.Data Availability StatementData are available from the corresponding author upon reasonable request.Author Contribution StatementAliyu Tetengi Ibrahim: Conceptualization, Methodology, Formal analysis, Writing – original draft. Ibrahim Hayatu Hassan: Conceptualization, Methodology, Formal analysis, Resources, Data curation, Writing – original draft. Mohammed Abdullahi: Validation, Formal analysis, Resources, Data curation, Writing – review & editing, Visualization. Abeer Rashad Mirdad: Conceptualization, Software, Validation, Investigation, Writing – original draft, Writing – review & editing, Supervision, Project administration. Muhammad I. Khan: Methodology, Validation, Investigation, Resources, Data curation, Writing – review & editing, Visualization, Project administration. Saeed Ali Bahaj: Conceptualization, Software, Writing – original draft. Fatima Khan Nayer: Investigation, Visualization, Supervision, Project administration.

Diabetes Mellitus (DM) is a condition where the body cannot regulate blood sugar due to improper insulin production or use, posing a significant global health burden. Traditional detection methods rely on clinical assessments and basic lab tests, but recent technological advancements suggest that Machine Learning (ML) algorithms can predict DM more effectively and efficiently. However, current ML models face challenges like feature redundancy, irrelevancy, and dataset imbalance, which can reduce accuracy and interpretability, ultimately affecting patient outcomes. This paper aims to address these challenges by developing an enhanced ML-based DM prediction model. The proposed model leverages an ensemble soft voting classifier, integrating the Random Forest, Logistic Regression, and Naïve Bayes algorithms. Feature importance determination is facilitated by the Gini Index Random Forest (GI-RF) algorithm. Additionally, three data imbalance handling techniques random oversampling (ROS), random undersampling (RUS), and the synthetic minority oversampling technique (SMOTE) are employed to mitigate biased model development. Initially, the GI-RF algorithm identifies the top 5 most informative features from the PIMA Indians Diabetes Dataset, originally comprising 8 features. Subsequently, the dataset is subjected to each of the three imbalance handling techniques. The performance of each model variation, incorporating different imbalance handling techniques is then extensively compared. The results demonstrate that ROS notably outperforms RUS and SMOTE across multiple metrics, including accuracy, F1 score, recall, and AUC. A comparative analysis with existing studies reveals the proposed method's notable improvements across all metrics, with increases of 5% in accuracy, 8% in precision, 13% in F1 score, 18% in recall, and 4% in AUC. This demonstrates the proposed model's overall robustness and effectiveness in predictive modeling, contributing to more accurate diagnosis and treatment of DM.

The COVID-19 pandemic accelerated new vaccine development. Limited safety data necessitated robust global safety surveillance to accurately identify and promptly communicate potential safety issues. The African Union Smart Safety Surveillance (AU-3S) program established the Joint Signal Management (JSM) group to support identification of potential vaccine safety concerns in five pilot countries (Ethiopia, Ghana, Kenya, Nigeria, South Africa), accounting for approximately 35

In medical diagnostics, brain tumor classification remains essential, as accurate and efficient models aid medical professionals in early detection and treatment planning. Deep learning methodologies for brain tumor classification have gained popularity due to their potential to deliver prompt and precise diagnostic results. This article proposes a novel classification technique that integrates the Xception model with a hybrid attention mechanism and progressive image resizing to enhance performance. The methodology is built on a combination of preprocessing techniques, transfer learning architecture reconstruction, and dynamic fine-tuning strategies. To optimize key hyper-parameters, this study employed the Dynamic Chaotic Whale Optimization Algorithm. Additionally, we developed a novel learning rate scheduler that dynamically adjusts the learning rate based on image size at each training phase, improving training efficiency and model adaptability. Batch sizes and layer freezing methods were also adjusted according to image size. We constructed an ensemble approach by preserving models trained on different image sizes and merging their results using weighted averaging, bagging, boosting, stacking, blending, and voting techniques. Our proposed method was evaluated on benchmark datasets achieving remarkable accuracies of 99.67%, 99.09%, and 99.67% compared to the classical algorithms.

Lung cancer is a fatal disease with a high mortality rate in patience. Early and accurate detection of this disease plays a crucial role in improving a patient's chances of survival. Traditional methods, such as Computed Tomography (CT) scans, have historically been employed for tumor localization and assessing cancer severity. However, these methods are time-consuming and may pose risks, including patient mortality before tumor identification. Given the challenges associated with lung cancer classification and the limitations of existing practices, there is a pressing need for innovative clinical data assessment tools to complement biopsies and offer a more precise characterization of the disease. Recent literature suggests the application of deep learning techniques for lung cancer detection. However, efficient training of deep learning models requires a substantial amount of data, and the availability of annotated data for lung cancer detection is often constrained, potentially resulting in overfitting or under-fitting issues and inaccurate predictions. To address these challenges, this dissertation proposes a novel deep learning architecture based on the hybridization of three pre-trained models with a support vector machine (SVM) optimized using the honey badger optimization algorithm (HBA). The process involves pre-processing the input images to ensure compatibility with pre-trained models, implementing augmentation techniques to expand the dataset and prevent overfitting, and employing a hybrid model consisting of AlexNet, VGG16, and GoogleNet for feature extraction. The extracted features are combined to generate hybrid features, which are then fed into a multi-class SVM optimized with HBA for classification. The proposed model was trained and tested using a lung cancer dataset from Iraq-Oncology Teaching Hospital and the National Centre for Cancer Diseases (IQ-OTH/NCCD), comprising 1190 images across three categories: normal, benign, and malignant. The model underwent validation and was compared with existing literature works. The results demonstrated superior performance, achieving an overall accuracy of 98% in accurately detecting different categories of lung cancer. This result demonstrates the capability of the proposed model compared to other existing models from the literature.

Cancer of the breast popularly known as breast cancer (BC) is the second and third utmost cause of mortality among women in Nigeria and globally, respectively. Biopsy histopathological images (BHI) have gained more attention for the early clinical diagnosis of BC. However, the clinical examination and diagnosis of BC histology images are subject to human error. Consequently, several computer-aided diagnoses (CAD) solutions have been presented to aid histopathologists with the automated classification of cancerous tumor cells on histological images. Deep convolutional neural networks (DCNN) have been utilized to build a sizable portion of the cutting-edge proposed solutions. However, due to the architectural structure of DCNN, which extracts features automatically along with training processes and is coupled with overlapping nucleic features on breast histology images (BHI), the existing solutions suffer from high computational utilization, extensive training time leading to longer convergence times, and reliance on available high-end system resources to build adequate BC classification solutions. In this paper, an enhanced shallow convolutional neural network (ES-CNN) has been proposed for multi-classification of BHI, aimed to improve classification performance and reduce training time across eight BC types and four magnifications in the BreakHis dataset. The research objectives were achieved in three ways. First, we designed the proposed network’s architecture, guided by magnification and patient dependencies. Secondly, we implemented a multi-classification model based on the proposed network, and, finally, two categories of experiments were conducted based on classification accuracy and computational utilization. The experimental results revealed that the proposed methods have minimal computational utilization and improved classification performance compared to the existing work. This research reports a multi-classification accuracy of 96

Feature Selection (FS) is critical for reducing the high dimensionality of data, which negatively impacts the classification performance of machine learning models. In the field of Intrusion Detection (ID), where datasets often consist of thousands of attributes and instances, the prevalent issue of data imbalance poses significant challenges, leading to bias in classification tasks. This highlights the pressing need for intelligent techniques to address these challenges effectively. Genetic Algorithm (GA), a widely used evolutionary optimization algorithm for FS, encounters limitations such as slow convergence and a tendency to settle prematurely on suboptimal solutions due to insufficient exploitation capability. These limitations can adversely affect overall performance. Additionally, conventional techniques like the Synthetic Minority Oversampling Technique, commonly employed to handle data imbalance, risk introducing noisy data points into the feature space. To overcome these issues, this study proposes an improved GA-based FS technique featuring an enhanced mutation operator to bolster its exploitation capabilities and deliver improved performance. Furthermore, Adaboost, a more promising machine learning algorithm, is suggested to effectively address data imbalance challenges. The performance of the proposed model was evaluated using benchmark datasets from the Security Laboratory Knowledge Discovery Dataset (NSL-KDD), employing five performance metrics, including accuracy, F1-score, recall, precision, and execution time. The results show that the proposed method outperforms existing techniques across all metrics while effectively tackling the challenges of high-dimensional data and imbalanced datasets, offering a reliable solution for Intrusion Detection.

Orofacial neoplasms in children and young adults may differ significantly from those observed in adults. Our aim was to describe the epidemiological characteristics of histologically diagnosed orofacial neoplasms among children and young adults in Nigeria. This was a multicenter cross-sectional study across geopolitical zones in Nigeria. Annual reports of clinical information and surgical biopsies submitted at the Oral Pathology Laboratory, clinic day registries, surgical day case registries and operative theatre registries were retrieved from January 2008 to March 2024. The relevant demographic data were obtained for each patient. The study subjects were categorized by age into children, adolescents and young adults. Tissue involvement was classified as soft tissue involvement, bony involvement or both soft tissue and bony involvement. The site and behaviour of the lesions were subdivided according to the ICD-10 codes. Statistical analysis was performed via the R programming language. A total of 1889 cases were observed during the period under review, with a mean age of 15 years. Cases were more common in females (52

Convolutional Neural Networks (ConvNets) have demonstrated impressive capabilities in image classification; however, the manual creation of these models is a labor-intensive and time-consuming endeavor due to their inherent complexity. This research introduces an innovative approach to Convolutional Neural Network (ConvNet) architecture generation through the utilization of the Symbiotic Organism Search ConvNet (SOS_ConvNet) algorithm. Leveraging the Symbiotic Organism Search optimization technique, SOS_ConvNet evolves ConvNet architectures tailored for diverse image classification tasks. The algorithm's distinctive feature lies in its ability to perform non-numeric computations, rendering it adaptable to intricate deep learning problems. To assess the effectiveness of SOS_ConvNet, experiments were conducted on diverse datasets, including MNIST, Fashion-MNIST, CIFAR-10, and the Breast Cancer dataset. Comparative analysis against existing models showcased the superior performance of SOS_ConvNet in terms of accuracy, error rate, and parameter efficiency. Notably, on the MNIST dataset, SOS_ConvNet achieved an impressive 0.31% error rate, while on Fashion-MNIST, it demonstrated a competitive 6.7% error rate, coupled with unparalleled parameter efficiency of 0.24 million parameters. The model excelled on CIFAR-10 and BreakHis datasets, yielding accuracies of 82.78% and 89.12%, respectively. Remarkably, the algorithm achieves remarkable accuracy while maintaining moderate model size.

Background: Vaccination strengthens the immune system by using the body’s intrinsic defense mechanisms to enhance resistance to notable or specific disease agents. A cross-sectional community-based study comparing the occurrence of Covid-like symptoms among individuals earlier vaccinated or unvaccinated against Covid-19 using all the four different vaccine types administered in Nigeria. Given that Sub-Sahara Africans were not included in the vaccine clinical trials before Emergency Use Authorization, this study could provide indirect evidence on the effectiveness or otherwise of the vaccines among Nigerians. Methods: A cross-sectional community-based study of prevalence of covid like symptoms reported in subjects who had received any of the Covid-19 vaccine types deployed in Nigeria compared with subjects who had not received any of the vaccines in the six geo political zones of the country. A multi-stage sampling technique was used to select 24 Local government Areas from all six (6) geopolitical zones of the country. Results: Lesser proportion of subjects who had received any of the COVID-19 vaccines reported experiencing symptoms such as headaches, muscle, joint aches, fatigue, fever, sore throat, cough, and chills in the preceding 3 months to the survey as compared to subjects who had not received any of the vaccines. These observed differences were statistically significant, (p<0.05). Conclusion: The COVID-19 vaccines would appear protective against the symptoms of interest based on these findings. The reduction in the proportions reporting symptoms for the different types of vaccine are invariably similar therefore vaccine recipients can be encouraged to receive any brand to curb the pandemic.

Patients diagnosed with head and neck squamous cell carcinoma (HNSCC), particularly those seropositive for human immunodeficiency virus (HIV), face a heightened risk of second primary malignancies (SPMs), with common regions being the head, neck, lung, and oesophagus. This risk amplifies the severity of their clinical condition, as these SPMs contribute significantly to the mortality rates in patients with HNSCC. We detail a case of a young woman, seropositive for HIV, who developed a second squamous cancer in the nasopharynx after achieving remission from her initial oropharyngeal squamous cell carcinoma through chemo-radiotherapy. This case study highlights the increased vulnerability of HIV-positive HNSCC patients to SPMs, with an observed association of HIV infection leading to a lower overall survival rate. As a result, we recommend long-term follow-up in HNSCC patients with HIV for early detection of SPMs. Our findings emphasize the importance of regular screening for HNSCC, particularly in people living with HIV, to ensure timely detection and treatment, which can significantly improve their prognosis.

Myelomeningocele can occur throughout the spinal axis but the area with the least common localization is the thoracic region which is uncommon. Rarely, these abnormalities coexist with a teratoma and only few reports of a teratoma arising from a myelomeningocele have been reported. Teratoma can also undergo malignant transformation to a non-germ cell malignancy. Here, we report an interesting case of a teratoma arising within thoracic myelomeningocele in an infant. A three month old girl born via vaginal delivery to two related Fulani parents presenting with upper back swelling noticed since birth. Pregnancy was claimed to be uneventful, mother never attended any antenatal clinic and delivery was at home. On examination patient was mildly dehydrated and malnourished. Status localis revealed an intact cystic midline upper back swelling measuring about 6x4cm, non-tender with no differential warmth. Radiographic evaluation was not done due to financial constraint. Grossly, three irregularly shaped firm to cystic to flabby tissues aggregating to 5x4x1.5cm, weighing 6g were received and microscopy revealed histologic features of a myelomeningocele with a teratoma having endodermal (respiratory epithelium, seromucinous glands) , ectodermal( skin) and mesodermal (cartilage) components.Teratoma is a germ cell neoplasm that contains differentiated elements derived from all three germ layers. The occurrence of a mature midline teratoma within a myelomeningocele is rare and what makes it even rarer is occurrence in the thoracic region. This can be confirmed by meticulous histopathologic evaluation.Myelomeningocele has variable prognosis based on localization and other accompanying abnormalities. Due to the risk of malignant transformation to non-germ cell tumours in teratoma, prompt histologic diagnosis and complete resection is important regardless of size of lesion and age of patient.

Visual Feature Learning (VFL) is a critical area of research in computer vision that involves the automatic extraction of features and patterns from images and videos. The applications of VFL are vast, including object detection and recognition, facial recognition, scene understanding, medical image analysis, and autonomous vehicles. In this paper, we propose to conduct extensive systematic literature review (SLR) on VFL based on deep learning algorithms. The paper conducted an SLR covering deep learning algorithms such as Convolutional Neural Networks (CNNs), Autoencoders, and Generative Adversarial Networks (GANs) including their variants. The review highlights the importance of VFL in computer vision and the limitations of traditional feature extraction techniques. Furthermore, it provides an in-depth analysis of the strengths and weaknesses of various deep learning algorithms for solving problems in VFL. The discussion of the applications of VFL provides an insight into the impact of VFL on various industries and domains. The review also analyzed the challenges faced by VFL, such as data scarcity and quality, overfitting, generalization, interpretability, and explainability. The discussion of future directions for VFL includes hybrid techniques, unsupervised feature learning, continual learning, attention-based models, and explainable AI. These techniques aim to address the challenges faced by VFL and improve the performance of the models. The systematic literature review concludes that VFL is a rapidly evolving field with the potential to transform many industries and domains. The review highlights the need for further research in VFL and emphasizes the importance of responsible use of VFL models in various applications. The review provides valuable insights for researchers and practitioners in the field of computer vision, who can use these insights to enhance their work and ensure the responsible use of VFL models.

BACKGROUND:This was a cross-sectional community-based survey to study the prevalence of serum antibodies against the severe acute respiratory syndrome coronavirus 1 (SARS-COV-1) and determine possible source of antibodies as to whether from vaccination or from natural infection as well as attempt to compare antibody levels in response to the different four types of vaccines administered in Nigeria. METHODS:A cross-sectional community-based study of the prevalence of serum antibodies against all four vaccine types used in Nigeria amongst a representative sample of people aged 18 years and above in the six geopolitical zones of the country using a multistage sampling technique covering 12 states of the country with two states being randomly selected from each geopolitical zone. High-throughput Roche electrochemiluminescence immunoassay system (Elecsys Anti-SARS-COV-1 Cobas) was used for qualitative and quantitative detection of antibodies to SARS-COV-1 in human plasma. RESULTS:There was no statistically significant difference between the proportions with seropositivity for both the vaccinated and the unvaccinated (P = 0.95). The nucleocapsid antibody (anti-Nc) titres were similar in both the vaccinated and the unvaccinated, whereas the Spike protein antibody (anti-S) titres were significantly higher amongst the vaccinated than amongst the unvaccinated. Antibody levels in subjects who received different vaccines were compared to provide information for policy. CONCLUSION:While only 45.9% of the subjects were reported to have been vaccinated, 98.7% of the subjects had had contact with the SARS-COV-1 as evidenced by the presence of nucleocapsid (NC) antibodies in their plasma. The 1.3% who had not been exposed to the virus, had spike protein antibodies which most likely resulted from vaccination in the absence of NC antibodies. Successive vaccination and booster doses either through heterogeneous or homologous vaccines increased antibody titres, and this stimulation of immune memory may offer greater protection against coronavirus disease 2019.

Metaheuristic algorithms are commonly used in solving complex and NP-hard optimization problems in various fields. These algorithms have become popular because of their ability to explore and exploit solutions in various problem domains. Honey Badger Algorithm (HBA) is a population-based metaheuristic optimization algorithm inspired by the dynamic hunting strategy of honey badgers, utilizing honey and digging-seeking techniques. Since its introduction in 2020, HBA has garnered widespread attention and has been applied across various domains. This review aims to comprehensively survey the improvement and application of HBA in solving various optimization problems. Additionally, the survey conducts a meta-analysis of the HBA's improvements, hybridization and application since its introduction. According to the result of the survey, 52 studies presented improved HBA using chaotic maps, levy flight mechanism, adaptive mechanisms, transfer functions, multi-objective mechanism and opposition based learning techniques, 20 studies presented a hybrid HBA with other metaheuristics algorithms and 101 studies uses the original HBA for solving various optimization problems. According to the survey, the wide acceptance of the HBA within the research community stems from its straightforwardness, ease of use, efficient computational time, accelerated convergence speed, high efficacy, and capability to address different kind of optimization issues, distinguishing it from the well-known optimization approches presented.

Background:Neurofibromas are the most common benign nerve sheath tumours occurring as solitary sporadic tumours or multiple Syndromic tumours associated with neurofibromatosis type 1(NF1). In Nigeria and West Africa, there is a paucity of literature and studies on neurofibromas. This study aims to analyse the histopathologic and immunohistochemical patterns of neurofibromas and determine the frequency, demographic and anatomic distributions. Methodology:The study was a hospital-based retrospective study, and the study population constituted all surgical specimens submitted for histological examination to the Department of Pathology between 1st January 2010 to 31st December 2019 reported as neurofibroma. Records were retrieved from the archives and subjected to histopathologic and immunohistochemical analysis following standard protocols. Collated data was analysed, slides were reviewed, and results were presented in frequency distribution tables and statistical charts. Results:A total of 125 cases were seen constituting 8.3% of all soft tissue tumours seen. Neurofibromas were more prevalent in females with a male-to-female ratio of 1:1.15. The age ranged between 2-70 years with a mean age of 25.38 years and the highest frequency of occurrence was in the second decade of life. The most frequent anatomic site of occurrence was the head and neck region. Most of the tumours 103 (82.4%) were sporadic while 22(17.6%) were Syndromic and associated with NF1. A malignant transformation of a pre-existing neurofibroma in an NF1 patient was seen. The most common histologic variant seen was the conventional variant. Ninety percent of these tumours showed SOX10 immunopositivity, 91% showed S100 immunopositivity and 95% showed CD34 immunopositivity. Calretinin expression was low showing 16%. No hot spots labeling index seen with Ki67 antibody. Conclusion:Neurofibromas are more common in females in our environment and the most frequent anatomic site of involvement is the head and neck region.

Skin cancer is the abnormal development of cells on the surface of the skin and is one of the most fatal diseases in humans. It usually appears in locations that are exposed to the sun, but can also appear in areas that are not regularly exposed to the sun. Due to the striking similarities between benign and malignant lesions, skin cancer detection remains a problem, even for expert dermatologists. Considering the inability of dermatologists to diagnose skin cancer accurately, a convolutional neural network (CNN) approach was used for skin cancer diagnosis. However, the CNN model requires a significant number of image datasets for better performance; thus, image augmentation and transfer learning techniques have been used in this study to boost the number of images and the performance of the model, because there are a limited number of medical images. This study proposes an ensemble transfer-learning-based model that can efficiently classify skin lesions into one of seven categories to aid dermatologists in skin cancer detection: (i) actinic keratoses, (ii) basal cell carcinoma, (iii) benign keratosis, (iv) dermatofibroma, (v) melanocytic nevi, (vi) melanoma, and (vii) vascular skin lesions. Five transfer learning models were used as the basis of the ensemble: MobileNet, EfficientNetV2B2, Xception, ResNext101, and DenseNet201. In addition to the stratified 10-fold cross-validation, the results of each individual model were fused to achieve greater classification accuracy. An annealing learning rate scheduler and test-time augmentation (TTA) were also used to increase the performance of the model during the training and testing stages. A total of 10,015 publicly available dermoscopy images from the HAM10000 (Human Against Machine) dataset, which contained samples from the seven common skin lesion categories, were used to train and evaluate the models. The proposed technique attained 94.49% accuracy on the dataset. These results suggest that this strategy can be useful for improving the accuracy of skin cancer classification. However, the weighted average of f1-score, recall, and precision were obtained to be 94.68%, 94.49%, and 95.07%, respectively.

This systematic review paper provides a comprehensive analysis of the recent advances in deep learning techniques for rice disease recognition. Rice is one of the most important crops in the world, providing food for more than half of the global population. However, rice diseases pose a major threat to rice production and can cause significant yield losses. In recent years, deep learning techniques have shown great potential in automating the process of rice disease recognition, which can help in early disease detection and management. This paper reviews the current trends in deep learning techniques for rice disease recognition, including various pre-processing and augmentation techniques, as well as popular deep learning models such as convolutional neural networks (CNNs) and their variants. The paper also provides an in-depth analysis of the different datasets used in the studies, along with their limitations and challenges. Furthermore, the paper discusses the future directions for research in this field, such as the need for larger and more diverse datasets, the development of novel deep learning architectures, and the integration of other data sources such as weather data and satellite imagery. The paper concludes by summarizing the key findings of the systematic review and highlighting the potential impact of deep learning techniques in rice disease recognition. In addition, the review provides a useful resource for researchers and practitioners in the field of agricultural technology and can help in the development of more accurate and efficient automated systems for rice disease detection and management.