Multiclass Blood Cell Classification Using Contourlet Transform and Metaheuristic-Optimized Deep Features with Clustering-Based Decision Making

Background: Accurate classification of human blood cells is crucial for diagnosing hematological disorders, including infections, inflammation, and leukemia. Manual examination of blood smears is widely used but is time-consuming, subjective, and prone to error. Automated approaches are needed to enhance diagnostic efficiency and reliability. Methods: This study proposes a hybrid automated classification framework that integrates deep learning with shape transformation techniques. Contourlet transform was employed for shape-based feature extraction, while a recurrent artificial neural network (RANN) was applied for deep feature learning. The African Vulture Optimization Algorithm (AVOA) was employed to optimize feature selection, and a clustering-based decision-making strategy was implemented for the final classification. Results: The proposed framework demonstrated high classification accuracy across five major blood cell types: lymphocytes (91%), monocytes (97%), eosinophils (94%), basophils (69%), and neutrophils (75%). The integration of contourlet transform and RANN improved feature representation, while AVOA enhanced classification robustness by optimizing feature subsets. Conclusion: The results indicate that the proposed hybrid model significantly improves diagnostic precision by combining shape-based and deep learning features with advanced optimization techniques. This framework shows potential for clinical translation as a reliable and efficient tool for automated hematology diagnostics.