A Metaheuristic-Optimized Framework for Automated Leukemia Detection from Peripheral Blood Smear Images

The peripheral blood smear (PBS) can be an important diagnostic tool for many diseases. Accurately detecting cells on a PBS faces many challenges, including image noise, non-uniform illumination, and poor contrast among cellular components. Newly developed an image processing system using a metaheuristic optimization technique to automate leukemia classification. First, a Kuan filter removes speckle noise while preserving important cellular structural features. Red Rose Optimization (RRO) finds the best preprocessing parameters to improve image quality. Frizzy Lizard Optimization (FLO) optimizes feature extraction. Gabor filters extract important textural and orientational characteristics from the PBS image. Prairie Dog Optimization (PDO) identifies the most informative features and removes redundant ones to reduce the feature set size. Finally, the selected features are used to classify diseases. This study compared PDO with Golden Jackal Optimization (GJO) and the Rice Optimization Algorithm (ROA) using the same testing methodology. The system's performance was evaluated based on accuracy, precision, sensitivity, specificity, F1-score, correlation coefficient, and area under the receiver operating characteristic curve (AUC-ROC). This methodology produced a small, effective feature set, improved classification reliability, and showed how metaheuristic optimization may aid in diagnosing leukemia through analysis of PBS images.