Explainable multimodal hematology analysis for white blood cell classification and attribute prediction
Abstract: Highlights•Proposes a CLIP-based multimodal model for WBC classification and attribute prediction.•Uses structured prompts to semantically align visual and morphological features.•Introduces adaptive weighting to address class imbalance in joint-task learning.•Achieves state-of-the-art results in WBC type and attribute prediction tasks.
External IDs:doi:10.1016/j.compbiomed.2025.110734
Loading