VGG16-Based Multiscale Feature Fusion for Robust Blood Cell Cancer Classification

Abstract

Leukemia, particularly acute lymphoblastic leukemia (ALL), is one of the most prevalent and fatal blood cancers, with more than 460,000 new cases and over 300,000 related deaths reported globally in 2021. Despite advances in deep learning, conventional convolutional neural network models applied to blood cell cancer classification often deliver suboptimal performance, with reported accuracies frequently below 90%. These limitations are due to difficulty in capturing subtle morphological variations, high similarity among malignant subtypes, and problems like class imbalance and noise in peripheral blood smear images. To overcome these difficulties, this study proposes a VGG16-based multiscale feature fusion (MFF) model that fuses the shallow and deep feature maps from different convolution blocks of the VGG16 backbone. Before classification, these feature maps are spatially aligned and concatenated into a single multiscale representation at a consistent resolution. The model was tested on the Blood Cell Cancer (ALL) dataset, which includes 3242 peripheral blood smear images that are classified into four categories: benign, malignant early Pre-B, malignant Pre-B, and malignant Pro-B. The experimental results showed that the proposed VGG16-MFF obtained 99.69% accuracy, precision, recall, and F1-score, which significantly outperformed baseline architectures such as ResNet50 (50.00%), ResNet101 (76.85%), VGG16 (88.27%), and VGG19 (87.34%). These results support the effectiveness of MFF in improving classification robustness, and future work may consider its extension to larger and more varied medical imaging datasets. 

Received: 30 August 2025 | Revised: 14 April 2026 | Accepted: 25 June 2026

Conflicts of Interest

The author declares that he has no conflicts of interest to this work. 

Data Availability Statement

The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/mohammadamireshraghi/blood-cell-cancer-all-4class. 

Author Contribution Statement

Simeon Yuda Prasetyo: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.