Recent metaheuristic algorithms for medical object localization using MSER detector in computer-aided diagnosis system

• Published in: Multimedia tools and applications Vol. 84; no. 15; pp. 14433 - 14487
• Main Authors: Ait Mehdi, Mohamed, Belattar, Khadidja, Souami, Feryel
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2025; Springer Nature B.V
• Subjects: Algorithms; Artificial intelligence; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Diagnosis; Exploitation; Heuristic methods; Image analysis; Localization; Medical imaging; Multimedia Information Systems; Object recognition; Rank tests; Special Purpose and Application-Based Systems; Track 6: Computer Vision for Multimedia Applications; Deep learning; Local binary pattern; Metaheuristic; Maximally stable extremal regions; Computer-aided diagnosis
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-19606-w

Object localization is a critical task in image analysis, often facilitated by artificial intelligence techniques. While the Maximally Stable Extremal Regions (MSER) detection algorithm is a popular choice for local detection, its exhaustive approaches can be algorithmically complex and prone to suboptimal results with improper parameter selection. Various metaheuristic algorithms have been proposed for medical object localization to address this. In this context, four contributions are presented. Firstly, recent metaheuristics, including the Slime Mould Algorithm (SMA), Marine Predators Algorithm (MPA), Heap-based Optimizer (HBO), and Gradient-based Optimizer (GBO), are adapted to tackle the MSER localization problem. These algorithms are rigorously evaluated across diverse medical image datasets using multiple metrics, with their performance statistically validated through the Friedman mean rank test. The second contribution introduces a novel objective function to improve the localization process. The third contribution involves a comparative analysis of the recent algorithms against seven standard metaheuristics specifically designed for MSER localization. Lastly, we present an improved computer-aided diagnosis system that integrates an SVM-based model with Local Binary Patterns (LBP) descriptors extracted from each MSER alongside DenseNet-121 features. Experimental results demonstrate that the HBO and SMA optimizers outperform conventional algorithms, showing a higher diversity, improved exploitation and exploration capabilities, a well-balanced exploration-exploitation trade-off, elevated fitness values, faster convergence speeds, and enhanced computational efficiency. Additionally, we integrate these findings into an improved computer-aided diagnosis system, yielding classification results surpassing state-of-the-art models. These findings highlight the potential advantages of this approach, especially in object detection applications.