HemaVision: EfficientNetV2 and CBAM-Based Explainable AI for Morphological White Blood Cell Classification and Clinical decision Support

• Novel Attention-Guided HemaVision Framework:
We propose HemaVision, an attention-enhanced deep
learning framework, integrating EfficientNetV2-B0 with
CBAM to precisely classify white blood cells into five
classes.
• High Classification Performance: The proposed frame-
work is able to accurately classify the white blood cells
with a validation accuracy of 99.04%, which shows a bet-
ter performance and reliability in automated classification
of white blood cells (WBCs).
• Grad-CAM Interpretability and Clinical Report Gen-
eration: To illustrate the model’s predictions visually,
we integrate Grad-CAM to the model that highlight the
image regions influencing the model’s predictions. After
that, we generate automated Clinical Decision Reports
that translate these visual localizations into valuable,
physiologically substantiated findings to support medical
decision-making.