Segmentation-free hybrid CNN–transformer CAD framework with handcrafted feature fusion for explainable Leukaemia diagnosis

Abstract

Diagnosing leukaemia from microscopic blood smear images is nevertheless challenging for traditional Computer-Aided Diagnosis (CAD) systems. Many current methods depend on segmentation and basic feature representations. This makes it hard to tell apart morphologically similar subtypes. This work presents an extended hybrid CAD framework to address these limitations by fusing handcrafted texture descriptors, deep CNN representations, Transformer-based contextual modelling and explainable AI. Multi-resolution handcrafted features using the Discrete Wavelet Transform (DWT), Grey-Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LBP) and statistical measures are combined with deep features from ResNet-50, DenseNet-121, and VGG-19. The feature combination is further refined using a Transformer encoder and SelectKBest feature selection. Lastly, classification is performed using Support Vector Machine (SVM), XGBoost, and ensemble models. The framework was authenticated on the ALL-IDB2 dataset and the large multi-subtype Leukaemia dataset consists of 20,000 images. On ALL-IDB2 dataset, the highest-performing configuration achieved 98.08% accuracy, 0.981 precision, 0.982 recall, 0.981 F1-score, and an AUC of 0.989. For the multi-subtype Leukaemia dataset, the model achieved 98.0% accuracy, 0.980 precision, 0.979 recall, 0.980 F1-score, and an AUC of 0.987. The experiments conducted using the ALL-IDB2 database using the reduced data set have shown that the proposed approach still manages to remain competitive despite using less training data sets, highlighting data efficiency and robustness rather than higher full-data accuracy. The outcomes from the Leukaemia dataset validate the standard performance under measured experimental conditions and should not be constructed as direct evidence of real-world clinical applicability. SHAP-based interpretability further highlights clinically relevant morphological features, improving transparency and diagnostic confidence. Diagrammatic representation of the Extended Hybrid CAD framework for leukaemia subtype classification.