Q-AVOA-net: A multi-domain, quantum-inspired feature-selection and deep-fuzzy framework for interpretable white blood cell classification

Accurate white blood cell (WBC) classification is crucial for hematological diagnostics, yet current automated systems struggle with data imbalance, limited generalization, and lack of explainability. This paper proposes Q-AVOA-Net, a quantum-inspired multi-domain framework that enhances WBC subtype classification accuracy, interpretability, and robustness. The model integrates: (i) Enhanced Contourlet Transform with learnable Gabor filters for multi-scale directional representation, (ii) Bi-LSTM Attention for hierarchical sequential learning, (iii) Quantum-inspired African Vulture Optimization Algorithm (Q-AVOA) for multi-objective feature selection, and (iv) an interpretable fuzzy decision layer for calibrated predictions. Extensive experiments on three benchmark datasets (Raabin, LISC, BCCD) demonstrate Q-AVOA-Net achieves 96.8% accuracy, 96.5% F1-score, and 0.991 AUC while reducing feature dimensionality by 38% and inference time by 22% compared to hybrid CNN-Transformer baselines. The framework also reduces Expected Calibration Error by ≈52%, significantly improving confidence estimation and clinical reliability. Q-AVOA-Net provides a robust, interpretable, and computationally efficient solution for next-generation hematological analysis and diagnostic decision support.