Efficient Acute Lymphoblastic Leukemia Classification Using GLCM and HSV Features with Reservoir Computing on Taleqani Hospital Dataset

Abstract

Detection of Acute Lymphoblastic Leukemia (ALL) from peripheral blood smear (PBS) images poses a major challenge in clinical diagnostics owing to intricate morphological variations and constrained computational resources in many healthcare facilities. This study proposes an efficient classification framework using Echo State Network (ESN) and Next-Generation Reservoir Computing (NG-RC), applied to a dataset of 3,256 PBS images from 89 patients. Feature extraction via Gray Level Co-occurrence Matrix (GLCM) and HSV histograms, combined with 10-fold cross-validation, yielded robust results. NG-RC achieved superior performance over ESN, with an average accuracy of 0.9917, precision of 0.9918, recall of 0.9917, F1-score of 0.9917, ROC-AUC of 0.9989, and remarkably low execution time of 1.88 seconds on CPU-only hardware. In additional benchmarking, NG-RC demonstrated competitive accuracy against high-end deep learning models trained on NVIDIA A100 GPU—DenseNet201 (accuracy 0.9978, execution time 176.5 s) and ResNet18 (accuracy 0.9966, execution time 54.6 s)—while outperforming most literature methods (e.g., YOLOv4 at 0.9600, Modified DCNN at 0.9769) and offering orders-of-magnitude faster inference without GPU dependency. These findings position NG-RC as a lightweight, resource-efficient alternative to deep learning, enabling reliable, rapid ALL detection in computer-aided diagnosis systems for resource-limited clinical settings.