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

Area of Science:

• Hematology
• Medical Imaging
• Artificial Intelligence

Background:

• Leukemia diagnosis is challenging due to morphological heterogeneity in optical microscopy.
• Accurate identification of leukemic cells is crucial for effective treatment.

Purpose of the Study:

• To develop a frequency-domain guided object detection framework for enhanced leukemia diagnosis.
• To improve the accuracy of distinguishing between different types of leukemia using microscopic images.

Main Methods:

• Utilized frequency-based image enhancement techniques on high-resolution bone marrow images.
• Implemented refined feature integration combining spatial and frequency information.
• Developed an object detection framework for leukemic cell classification.

Main Results:

• Achieved high precision in distinguishing acute lymphocytic leukemia (ALL) with an average precision of 89.7%.
• Achieved high precision in distinguishing chronic lymphocytic leukemia (CLL) with an average precision of 95.6%.
• Demonstrated the framework's ability to capture critical fine-grained details and semantic patterns.

Conclusions:

• The proposed AI framework significantly enhances diagnostic accuracy for leukemia classification.
• Integrating artificial intelligence with optical microscopy offers a valuable tool for hematologic malignancy diagnosis.
• Frequency-domain analysis combined with object detection shows promise for improving leukemia identification.