A novel feature reconstruction method for bone marrow cell classification

Huixiang Zhi

doi:10.1016/j.engappai.2026.113851

A novel feature reconstruction method for bone marrow cell classification

2026-01-21

0

PRE

AI

DOI：10.1016/j.engappai.2026.113851

Original

Original request for help

Save

Abstract

En 中文

Bone Marrow Cells (BMCs) play a crucial role in human health, particularly in hematopoiesis and immune function. Assessing BMC health is essential for the early detection and treatment of hematopoietic diseases. However, traditional morphological analysis heavily depends on the experience and expertise of practitioners. Meanwhile, Convolutional Neural Networks (CNNs) often suffer from performance degradation due to redundant information present in BMC images. To address this issue, we propose a novel module, the Reduced Redundancy Block (RR-Block), which enhances feature extraction by preserving high-frequency information while filtering out less significant features. Furthermore, the RR-Block is a plug-and-play module that can be seamlessly integrated into existing CNN architectures. Experimental results demonstrate that replacing the standard 3 × 3 convolutional layers in Residual Networks (ResNet) with the RR-Block significantly improves classification performance. Specifically, on the Peripheral Blood Cell (PBC) dataset, which comprises eight cell types, our approach achieved an accuracy of 98.54 % and a Cohen's Kappa coefficient (Kappa) of 99.37 %. Similarly, on the Munich Leukemia Laboratory dataset, which contains 21 cell classes, our model attained an accuracy of 91.69 % and a Kappa coefficient of 89.14 %. These results highlight the effectiveness of our model architecture, outperforming several state-of-the-art methods.

A novel feature reconstruction method for bone marrow cell classification

Abstract

Journal

Organization