Efficient diagnosis of hematologic malignancies using bone marrow microscopic images: A method based on MultiPathGAN...

Area of Science:

• Computational pathology leverages artificial intelligence for disease diagnosis.
• Deep learning models are advancing medical image analysis in hematology.
• Biomedical imaging integrates machine learning for clinical applications.

Background:

• Hematologic malignancies pose significant health risks, necessitating early detection.
• Bone marrow smear examination is crucial but traditionally labor-intensive.
• Current diagnostic methods for blood cancers require significant time and expertise.

Purpose of the Study:

• To develop an efficient AI-driven method for diagnosing hematologic malignancies.
• To enable direct diagnosis from bone marrow microscopic images.
• To overcome the time and labor constraints of conventional diagnostic techniques.

Main Methods:

• A deep learning framework utilizing the MobileViTv2 hybrid model was developed.
• Stain normalization using MultiPathGAN augmented a dataset of 2033 bone marrow images.
• The model was trained and validated on images from 61 individuals.

Main Results:

• MobileViTv2 achieved 94.28% accuracy on the test set, with high accuracy for multiple myeloma (98%), acute lymphocytic leukemia (96%), and lymphoma (96%).
• Patient-level prediction accuracy reached 96.72%.
• The model outperformed CNN and ViT models with fewer parameters and showed excellent performance on public datasets.

Conclusions:

• The developed AI framework effectively diagnoses hematologic malignancies from bone marrow images.
• The method is robust across different stain styles, offering efficient diagnostic capabilities.
• This approach promises to streamline the diagnosis of critical blood cancers.