Early and accurate detection of melanoma remains a critical challenge in medical imaging. Convolutional Neural Networks (CNNs) have demonstrated superior classification performance, often surpassing dermatologists in diagnostic accuracy. However, optimizing CNN architectures for clinical applications requires careful consideration of performance metrics that reflect both overall accuracy and Recall (sensitivity) to malignant cases. This article explores 45 recent research contributions addressing melanoma classification, offering a critical overview of current trends and techniques. Based on these findings, we propose a novel weighted multi-objective function (MOF) that balances key performance metrics according to clinical needs, with particular emphasis on Accuracy for overall correctness, Recall for reducing false negatives, and AUC for enhancing class discrimination in melanoma detection, thus ensuring both general reliability and high sensitivity in detecting melanoma. The Grey Wolf Optimizer (GWO), a population-based metaheuristic known for its effective exploration of high-dimensional search spaces, is employed for CNN hyperparameter tuning. Experimental results demonstrate that the proposed approach enhances model performance across key clinical metrics, obtaining a MOF score of 0.9032, with Accuracy, Recall, and AUC values of 0.9025%, 0.9101% and 0.8925, respectively. This confirms the effectiveness of the strategy in achieving both high diagnostic accuracy and sensitivity, offering a robust and scalable strategy for medical image classification in computer-aided diagnosis systems.
Adaptive decisions are often based on algorithm-specific feedback or simple indicators that provide limited information about the evolving search dynamics. This work investigates explicit pattern-based analysis of the search trajectory as a source of feedback for adaptive configuration control. A trajectory-driven framework is proposed in which Derivative Dynamic Time Warping (DDTW) analyzes the recent best-so-far fitness trajectory against reference patterns representing sustained progress and lack of progress. The resulting information regulates switching between predefined exploration- and exploitation-oriented parameter configurations. Two control strategies are introduced: Binary-Simple, which reacts directly to trajectory evidence associated with a plateau, and Binary-Complex, which incorporates multiple search conditions, temporal persistence, and asymmetric switching. The framework is evaluated on 27 Multidimensional Knapsack Problem instances using BPSO, GA, BGWO, and BDE, with 31 independent runs per condition. Results show that no fixed configuration consistently dominates across metaheuristics and instances, while the effects of trajectory-driven adaptation depend on the underlying algorithm. Across 108 metaheuristic–instance conditions, Binary-Complex achieves the best mean rank (2.13). The Friedman test detects an overall difference among strategies (); Bonferroni-Dunn post-hoc comparisons favor Binary-Complex over both fixed configurations, but do not establish a difference from Binary-Simple. Moreover, a common trajectory-analysis parameterization produces different switching behaviors across the evaluated metaheuristics. These findings support explicit search-trajectory pattern analysis as a viable feedback mechanism for adaptive configuration control across different population-based metaheuristics.
Background: Viral Pneumonia and Tuberculosis continue to represent a significant burden on global public health, relying heavily on chest X-rays for screening and diagnosis. Although deep learning systems offer promising diagnostic support, the traditional manual tuning of hyperparameters for Convolutional Neural Networks is often inefficient and computationally expensive, frequently resulting in suboptimal or overly heavy architectures. Methods: To address these challenges, this study proposes a hybrid framework that employs metaheuristic algorithms, specifically the Whale Optimization Algorithm, Grey Wolf Optimizer, and Cuckoo Search to automatically optimize the architecture and training parameters of a custom neural network for the multi-class classification of Normal, Viral Pneumonia, and Tuberculosis cases. The proposed approach was evaluated using a rigorous stratified k-fold cross-validation protocol on a balanced, multi-source dataset. Results: The experimental results demonstrate that the model optimized by the Whale Optimization Algorithm statistically outperforms manually configured baselines, achieving the highest diagnostic accuracy and specificity. Furthermore, a critical finding of this research is the substantial improvement in computational efficiency; the automated optimization reduced the computational load by approximately 74% and the storage requirements by 63%, making the model viable for deployment in resource-constrained environments. Conclusions: Finally, to ensure clinical reliability, the decision-making process was validated using Gradient-weighted Class Activation Mapping, which confirmed that the network successfully learns to identify clinically relevant pulmonary structures while ignoring confounding artifacts.