An Improved LNN–ACO Framework with Optimal Feature Selection for Breast Tumor Detection

Conventional image processing approaches, such as segmentation and feature extraction, may not perform effectively across different histopathological images due to highly variable structures. In addition, histopathological images are high-dimensional with gigapixel slides. However, the dynamic nature of cellular structures, temporal changes, and variations in nuclear size and shape present substantial challenges in terms of complex structure and staining variability. In addition, training deep CNN models from scratch may lead to overfitting. To address these issues, this study combines Liquid Neural Networks (LNN) with Ant Colony Optimization (ACO) (LNN-ACO) for breast tumor detection, reducing redundant features and fine-tuning model parameters, leading to a more discriminative feature space. Experimentation with the BreakHis dataset showed that the proposed LNN-ACO model achieved 93.75% accuracy, outperforming other techniques.