Attention-Aware Residual Network Based Manifold Learning for White Blood Cells Classification

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning

Background:

• Accurate classification of six white blood cell (WBC) types is crucial for leukemia diagnosis.
• Manual WBC classification is time-consuming and requires significant clinical expertise.
• Existing automated methods may lack efficiency or sufficient accuracy.

Purpose of the Study:

• To develop an efficient and automatic method for WBC classification to aid leukemia diagnosis.
• To introduce the Attention-aware Residual Network based Manifold Learning (ARML) model.
• To improve upon the limitations of existing manual and automated classification techniques.

Main Methods:

• The ARML model utilizes adaptive attention-aware residual learning to extract relevant image-level features.
• It incorporates both first- and second-order features, encoded via Gaussian embedding into a Riemannian manifold.
• The model is trained end-to-end with iterative optimization on a dataset of 10,800 WBC images.

Main Results:

• ARML achieved an average classification accuracy of 0.953 on a test set of 1,800 images.
• The model outperformed existing state-of-the-art methods while using fewer trainable parameters.
• Ablation studies confirmed the effectiveness of both manifold learning and attention-aware components.

Conclusions:

• The ARML model offers a robust and efficient solution for automated WBC classification.
• Its ability to learn distinguishable features using manifold learning enhances diagnostic potential.
• ARML presents a clinically feasible approach for leukemia diagnosis, reducing reliance on manual expertise.