Frequency-Domain Object Detection Network for Leukemia Diagnosis in Bone Marrow Microscopy.

Leukemia remains a prevalent hematologic malignancy, and its morphological heterogeneity presents challenges for reliable identification under optical microscopy. To address this, we propose a frequency-domain guided object detection framework to assist leukemia diagnosis using high-resolution bone marrow microscopic images. Specifically, we leverage frequency-based image enhancement and refined feature integration to improve the detection and classification of leukemic cells. By combining spatial and frequency information, our approach captures both fine-grained details and broader semantic patterns critical for accurate diagnosis. We validated our method on clinical microscopic images, achieving high precision in distinguishing acute lymphocytic leukemia (ALL) and chronic lymphocytic leukemia (CLL), with average precision rates of 89.7% and 95.6%, respectively. Our findings demonstrate the value of integrating artificial intelligence with optical microscopy for enhanced diagnostic accuracy in leukemia classification.