High-efficiency spatially guided learning network for lymphoblastic leukemia detection in bone marrow microscopy...

Area of Science:

• Hematology
• Medical Imaging
• Artificial Intelligence

Background:

• Leukemia diagnosis relies on subjective analysis of bone marrow smears, which is time-consuming and complex.
• Automated leukemia detection is hindered by limited datasets and challenges in whole slide image analysis, including cell morphology variations and image quality issues.

Purpose of the Study:

• To develop a novel dataset and an automated diagnostic method for accurate and rapid detection of lymphocytic leukemia.
• To establish a new benchmark for lymphocytic leukemia detection using high-quality microscopic images.

Main Methods:

• Construction of a new dataset with 1794 high-quality microscopic images.
• Development of a spatially-guided learning framework (SGLNet) for automated whole slide analysis.
• Implementation of scale-aware fusion, small object enhancement, and an efficient IoU loss function to improve detection accuracy.

Main Results:

• SGLNet achieved high mean average precision scores: 95.9% for acute lymphoblastic leukemia and 98.6% for chronic lymphocytic leukemia.
• The developed method effectively addresses challenges like morphological similarity and complex backgrounds in bone marrow smears.

Conclusions:

• SGLNet demonstrates high efficiency and accuracy in identifying lymphoblastic leukemia cells.
• The method significantly enhances large-scale clinical diagnosis and supports personalized treatment planning for leukemia patients.