Weakly Supervised Ternary Stream Data Augmentation Fine-Grained Classification Network for Identifying Acute...

Area of Science:

• Hematology
• Medical Imaging
• Artificial Intelligence

Background:

• Acute lymphoblastic leukemia (ALL) requires timely diagnosis via blood smear microscopy.
• Manual screening is labor-intensive and prone to errors.
• Existing AI models struggle with limited data and overfitting.

Purpose of the Study:

• To develop an AI-driven method for accurate and early diagnosis of ALL.
• To improve the performance of deep learning models for leukemia detection.
• To address challenges in distinguishing lymphoblasts from other lymphocytes.

Main Methods:

• A ternary stream-driven weakly supervised data augmentation classification network (WT-DFN) was developed.
• Attention maps were generated to identify key regions in microscopic images.
• Attention cropping and erasing techniques were used to extract fine-grained features.

Main Results:

• The WT-DFN improved classification accuracy by focusing on detailed object features and discriminative parts.
• The method effectively handled high intraclass and low interclass variances in lymphocyte images.
• Superior performance was achieved on both public and clinical datasets compared to other methods.

Conclusions:

• The proposed WT-DFN offers a robust solution for accurate ALL diagnosis using microscopic images.
• Weakly supervised attention mechanisms enhance feature extraction for improved classification.
• This approach holds promise for overcoming limitations in current AI-based diagnostic tools for leukemia.