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

Reliable white blood cell (WBC) classification from peripheral blood smear microscopy is hindered by stain variability, domain shift, class imbalance, and limited interpretability. We propose Q-AVOA-Net, a multi-domain and clinician-oriented framework integrating Enhanced Contourlet Transform and a Learnable Gabor Filter Bank for structure-texture encoding, Bi-LSTM attention for dependency modeling, Q-AVOA for multi-objective feature selection, and an interpretable fuzzy decision layer that outputs class-wise membership evidence. Evaluated on Raabin-WBC, LISC, and BCCD with cross-dataset and robustness testing, Q-AVOA-Net achieves 97.2 ± 0.2% accuracy on the integrated evaluation framework (AUC 0.988) and improves confidence reliability (ECE 0.018). The selected representation is compact (1280 features) while maintaining efficient inference (20.7 ms/image). Together, these results support robust, calibrated, and interpretable WBC classification for microscopy-based decision support.