Work

Hybrid CNN‐GRU Architecture With Feature Fusion and Uncertainty Quantification for Acute Lymphoblastic Leukemia Detection

Year: 2026

Type: article

Abstract: Acute lymphoblastic leukemia (ALL) is the most malignant form of leukemia. This form of leukemia is considered the most prevalent kind of cancer in both adults and children. Leukemia is typically identified by analyzing blood and bone marrow smears under a microscope. Advanced cytochemical testing can also be performed to confirm and categorize leukemia. However, these approaches are expensive, time‐consuming, and dependent on the skill and knowledge of the relevant specialists. In recent decades, deep learning with convolutional neural networks (CNNs) has generated sophisticated methods for image classification by analyzing microscopic smear images to detect the presence of leukemic cells. These techniques are easy to use, quick, affordable, and free from expert bias. Nevertheless, the majority of these approaches are unable to quantify the uncertainty in their results, which can have devastating consequences. This research implemented hybrid models that merged InceptionV3‐GRU, EfficientNetB3‐GRU, and MobileNetV2‐GRU for detecting and classifying ALL. Following this, to identify an optimal set of hyperparameters and enhance the model′s performance, Bayesian optimization is utilized. Then, we use a deep ensemble uncertainty quantification approach to deal with uncertainty during ALL image classification. The outcomes of three different hybrid deep learning models were then aggregated at the score level using the sum rule after they had been trained on the two publicly available ALL patient blood sample datasets: ALL‐IDB1 and ALL‐IDB2. Herein, a parallel architecture was considered, which offers a high degree of confidence in discriminating between ALL and not ALL. The proposed system has managed to correctly and accurately diagnose the ALL patients, respectively, with a detection accuracy rate of 100% for the ALL‐IDB1 dataset, 98.07% for the ALL‐IDB2 dataset, and 98.64% for the combined dataset using our proposed method.

Source: Journal of Engineering

DOI: 10.1155/je/7765920

Authors: Maksuda Akter, Rabea Khatun, Md. Manowarul Islam, Md. Ashraf Uddin, Md. Zulfiker Mahmud

Institutions: Jagannath University, Primeasia University, East West University

Language: English

Topic: Digital Imaging for Blood Diseases

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

SDG: Good health and well-being

Open Access status: gold

Estimated APC paid: $1,400