A Lightweight Plasma Cell Detection Framework with Knowledge Distillation Based Feature Refinement for Multiple Myeloma Diagnosis

This research proposes a lightweight YOLOv8n-based framework enhanced with P3 feature refinement to improve detection performance while maintaining model efficiency. Feature-level knowledge distillation is further investigated to assess the effectiveness of transferring information from a larger teacher model to the lightweight detector. The study also incorporates error analysis and explainable AI to provide deeper insight into model performance and limitations.